research paper about graph

• Open access
• Published: 16 January 2024

A review of graph neural networks: concepts, architectures, techniques, challenges, datasets, applications, and future directions

• Bharti Khemani 1 ,
• Shruti Patil 2 ,
• Ketan Kotecha 2 &
• Sudeep Tanwar 3  

Journal of Big Data volume  11 , Article number:  18 ( 2024 ) Cite this article

5933 Accesses

1 Citations

Metrics details

Deep learning has seen significant growth recently and is now applied to a wide range of conventional use cases, including graphs. Graph data provides relational information between elements and is a standard data format for various machine learning and deep learning tasks. Models that can learn from such inputs are essential for working with graph data effectively. This paper identifies nodes and edges within specific applications, such as text, entities, and relations, to create graph structures. Different applications may require various graph neural network (GNN) models. GNNs facilitate the exchange of information between nodes in a graph, enabling them to understand dependencies within the nodes and edges. The paper delves into specific GNN models like graph convolution networks (GCNs), GraphSAGE, and graph attention networks (GATs), which are widely used in various applications today. It also discusses the message-passing mechanism employed by GNN models and examines the strengths and limitations of these models in different domains. Furthermore, the paper explores the diverse applications of GNNs, the datasets commonly used with them, and the Python libraries that support GNN models. It offers an extensive overview of the landscape of GNN research and its practical implementations.

Introduction

Graph Neural Networks (GNNs) have emerged as a transformative paradigm in machine learning and artificial intelligence. The ubiquitous presence of interconnected data in various domains, from social networks and biology to recommendation systems and cybersecurity, has fueled the rapid evolution of GNNs. These networks have displayed remarkable capabilities in modeling and understanding complex relationships, making them pivotal in solving real-world problems that traditional machine-learning models struggle to address. GNNs’ unique ability to capture intricate structural information inherent in graph-structured data is significant. This information often manifests as dependencies, connections, and contextual relationships essential for making informed predictions and decisions. Consequently, GNNs have been adopted and extended across various applications, redefining what is possible in machine learning.

In this comprehensive review, we embark on a journey through the multifaceted landscape of Graph Neural Networks, encompassing an array of critical aspects. Our study is motivated by the ever-increasing literature and diverse perspectives within the field. We aim to provide researchers, practitioners, and students with a holistic understanding of GNNs, serving as an invaluable resource to navigate the intricacies of this dynamic field. The scope of this review is extensive, covering fundamental concepts that underlie GNNs, various architectural designs, techniques for training and inference, prevalent challenges and limitations, the diversity of datasets utilized, and practical applications spanning a myriad of domains. Furthermore, we delve into the intriguing future directions that GNN research will likely explore, shedding light on the exciting possibilities.

In recent years, deep learning (DL) has been called the gold standard in machine learning (ML). It has also steadily evolved into the most widely used computational technique in ML, producing excellent results on various challenging cognitive tasks, sometimes even matching or outperforming human ability. One benefit of DL is its capacity to learn enormous amounts of data [ 1 ]. GNN variations such as graph convolutional networks (GCNs), graph attention networks (GATs), and GraphSAGE have shown groundbreaking performance on various deep learning tasks in recent years [ 2 ].

A graph is a data structure that consists of nodes (also called vertices) and edges. Mathematically, it is defined as G = (V, E), where V denotes the nodes and E denotes the edges. Edges in a graph can be directed or undirected based on whether directional dependencies exist between nodes. A graph can represent various data structures, such as social networks, knowledge graphs, and protein–protein interaction networks. Graphs are non-Euclidean spaces, meaning that the distance between two nodes in a graph is not necessarily equal to the distance between their coordinates in an Euclidean space. This makes applying traditional neural networks to graph data difficult, as they are typically designed for Euclidean data.

Graph neural networks (GNNs) are a type of deep learning model that can be used to learn from graph data. GNNs use a message-passing mechanism to aggregate information from neighboring nodes, allowing them to capture the complex relationships in graphs. GNNs are effective for various tasks, including node classification, link prediction, and clustering.

Organization of paper

The paper is organized as follows:

The primary focus of this research is to comprehensively examine Concepts, Architectures, Techniques, Challenges, Datasets, Applications, and Future Directions within the realm of Graph Neural Networks.

The paper delves into the Evolution and Motivation behind the development of Graph Neural Networks, including an analysis of the growth of publication counts over the years.

It provides an in-depth exploration of the Message Passing Mechanism used in Graph Neural Networks.

The study presents a concise summary of GNN learning styles and GNN models, complemented by an extensive literature review.

The paper thoroughly analyzes the Advantages and Limitations of GNN models when applied to various domains.

It offers a comprehensive overview of GNN applications, the datasets commonly used with GNNs, and the array of Python libraries that support GNN models.

In addition, the research identifies and addresses specific research gaps, outlining potential future directions in the field.

" Introduction " section describes the Introduction to GNN. " Background study " section provides background details in terms of the Evolution of GNN. " Research motivation " section describes the research motivation behind GNN. Section IV describes the GNN message-passing mechanism and the detailed description of GNN with its Structure, Learning Styles, and Types of tasks. " GNN Models and Comparative Analysis of GNN Models " section describes the GNN models with their literature review details and comparative study of different GNN models. " Graph Neural Network Applications " section describes the application of GNN. And finally, future direction and conclusions are defined in " Future Directions of Graph Neural Network " and " Conclusions " sections, respectively. Figure  1 gives the overall structure of the paper.

The overall structure of the paper

Background study

As shown in Fig.  2 below, the evolution of GNNs started in 2005. For the past 5 years, research in this area has been going into great detail. Neural graph networks are being used by practically all researchers in fields such as NLP, computer vision, and healthcare.

Year-wise publication count of GNN (2005–2022)

Graph neural network research evolution

Graph neural networks (GNNs) were first proposed in 2005, but only recently have they begun to gain traction. GNNs were first introduced by Gori [2005] and Scarselli [2004, 2009]. A node's attributes and connected nodes in the graph serve as its natural definitions. A GNN aims to learn a state embedding h v ε R s that encapsulates each node's neighborhood data. The distribution of the expected node label is one example of the output. An s-dimension vector of node v, the state embedding h v , can be utilized to generate an output O v , such as the anticipated distribution node name. The predicted node label (O v ) distribution is created using the state embedding h v [ 30 ]. Thomas Kipf and Max Welling introduced the convolutional graph network (GCN) in 2017. A GCN layer defines a localized spectral filter's first-order approximation on graphs. GCNs can be thought of as convolutional neural networks that have been expanded to handle graph-structured data.

Graph neural network evolution

As shown in Fig.  3 below, research on graph neural networks (GNNs) began in 2005 and is still ongoing. GNNs can define a broader class of graphs that can be used for node-focused tasks, edge-focused tasks, graph-focused tasks, and many other applications. In 2005, Marco Gori introduced the concept of GNNs and defined recursive neural networks extended by GNNs [ 4 ]. Franco Scarselli also explained the concepts for ranking web pages with the help of GNNs in 2005 [ 5 ]. In 2006, Swapnil Gandhi and Anand Padmanabha Iyer of Microsoft Research introduced distributed deep graph learning at scale, which defines a deep graph neural network [ 6 ]. They explained new concepts such as GCN, GAT, etc. [ 1 ]. Pucci and Gori used GNN concepts in the recommendation system.

Graph Neural Network Evolution

2007 Chun Guang Li, Jun Guo, and Hong-gang Zhang used a semi-supervised learning concept with GNNs [ 7 ]. They proposed a pruning method to enhance the basic GNN to resolve the problem of choosing the neighborhood scale parameter. In 2008, Ziwei Zhang introduced a new concept of Eigen-GNN [ 8 ], which works well with several GNN models. In 2009, Abhijeet V introduced the GNN concept in fuzzy networks [ 9 ], proposing a granular reflex fuzzy min–max neural network for classification. In 2010, DK Chaturvedi explained the concept of GNN for soft computing techniques [ 10 ]. Also, in 2010, GNNs were widely used in many applications. In 2010, Tanzima Hashem discussed privacy-preserving group nearest neighbor queries [ 11 ]. The first initiative to use GNNs for knowledge graph embedding is R-GCN, which suggests a relation-specific transformation in the message-passing phases to deal with various relations.

Similarly, from 2011 to 2017, all authors surveyed a new concept of GNNs, and the survey linearly increased from 2018 onwards. Our paper shows that GNN models such as GCN, GAT, RGCN, and so on are helpful [ 12 ].

Literature review

In the Table  1 describe the literature survey on graph neural networks, including the application area, the data set used, the model applied, and performance evaluation. The literature is from the years 2018 to 2023.

Research motivation

We employ grid data structures for normalization of image inputs, typically using an n*n-sized filter. The result is computed by applying an aggregation or maximum function. This process works effectively due to the inherent fixed structure of images. We position the grid over the image, move the filter across it, and derive the output vector as depicted on the left side of Fig.  4 . In contrast, this approach is unsuitable when working with graphs. Graphs lack a predefined structure for data storage, and there is no inherent knowledge of node-to-neighbor relationships, as illustrated on the right side of Fig.  4 . To overcome this limitation, we focus on graph convolution.

CNN In Euclidean Space (Left), GNN In Euclidean Space (Right)

In the context of GCNs, convolutional operations are adapted to handle graphs’ irregular and non-grid-like structures. These operations typically involve aggregating information from neighboring nodes to update the features of a central node. CNNs are primarily used for grid-like data structures, such as images. They are well-suited for tasks where spatial relationships between neighboring elements are crucial, as in image processing. CNNs use convolutional layers to scan small local receptive fields and learn hierarchical representations. GNNs are designed for graph-structured data, where edges connect entities (nodes). Graphs can represent various relationships, such as social networks, citation networks, or molecular structures. GNNs perform operations that aggregate information from neighboring nodes to update the features of a central node. CNNs excel in processing grid-like data with spatial dependencies; GNNs are designed to handle graph-structured data with complex relationships and dependencies between entities.

Limitation of CNN over GNN

Graph Neural Networks (GNNs) draw inspiration from Convolutional Neural Networks (CNNs). Before delving into the intricacies of GNNs, it is essential to understand why Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) may not suffice for effectively handling data structured as graphs. As illustrated in Fig.  5 , Convolutional Neural Networks (CNNs) are designed for data that exhibits a grid structure, such as images. Conversely, Recurrent Neural Networks (RNNs) are tailored to sequences, like text.

Convolution can be performed if the input is an image using an n*n mask (Left). Convolution can't be achieved if the input is a graph using an n*n mask. (Right)

Typically, we use arrays for storage when working with text data. Likewise, for image data, matrices are the preferred choice. However, as depicted in Fig.  5 , arrays and matrices fall short when dealing with graph data. In the case of graphs, we require a specialized technique known as Graph Convolution. This approach enables deep neural networks to handle graph-structured data directly, leading to a graph neural network.

Fig. 5 illustrates that we can employ masking techniques and apply filtering operations to transform the data into vector form when we have images. Conversely, traditional masking methods are not applicable when dealing with graph data as input, as shown in the right image.

Graph neural network

Graph Neural Networks, or GNNs, are a class of neural networks tailored for handling data organized in graph structures. Graphs are mathematical representations of nodes connected by edges, making them ideal for modeling relationships and dependencies in complex systems. GNNs have the inherent ability to learn and reason about graph-structured data, enabling diverse applications. In this section, we first explained the passing mechanism of GNN (" Message Passing Mechanism in Graph Neural Network Section "), then described graphs related to the structure of graphs, graph types, and graph learning styles (" Description of GNN Taxonomy " Section).

Message passing mechanism in graph neural network

Graph symmetries are maintained using a GNN, an optimizable transformation on all graph properties (nodes, edges, and global context) (permutation invariances). Because a GNN does not alter the connectivity of the input graph, the output may be characterized using the same adjacency list and feature vector count as the input graph. However, the output graph has updated embeddings because the GNN modified each node, edge, and global-context representation.

In Fig. 6 , circles are nodes, and empty boxes show aggregation of neighbor/adjacent nodes. The model aggregates messages from A's local graph neighbors (i.e., B, C, and D). In turn, the messages coming from neighbors are based on information aggregated from their respective neighborhoods, and so on. This visualization shows a two-layer version of a message-passing model. Notice that the computation graph of the GNN forms a tree structure by unfolding the neighborhood around the target node [ 17 ]. Graph neural networks (GNNs) are neural models that capture the dependence of graphs via message passing between the nodes of graphs [ 30 ].

How a single node aggregates messages from its adjacent neighbor nodes

The message-passing mechanism of Graph Neural Networks is shown in Fig. 7 . In this, we take an input graph with a set of node features X ε R d ⇥ |V| and Use this knowledge to produce node embeddings z u . However, we will also review how the GNN framework may embed subgraphs and whole graphs.

Message passing mechanism in GNN

At each iteration, each node collects information from the neighborhood around it. Each node embedding has more data from distant reaches of the graph as these iterations progress. After the first iteration (k = 1), each node embedding expressly retains information from its 1-hop neighborhood, which may be accessed via a path in the length graph 1. [ 31 ]. After the second iteration (k = 2), each node embedding contains data from its 2-hop neighborhood; generally, after k iterations, each node embedding includes data from its k-hop setting. The kind of “information” this message passes consists of two main parts: structural information about the graph (i.e., degree of nodes, etc.), and the other is feature-based.

In the message-passing mechanism of a neural network, each node has its message stored in the form of feature vectors, and each time, the neighbor updates the information in the form of the feature vector [ 1 ]. This process aggregates the information, which means the grey node is connected to the blue node. Both features are aggregated and form new feature vectors by updating the values to include the new message.

Equations  4.1 and 4.2 shows that h denotes the message, u represents the node number, and k indicates the iteration number. Where AGGREGATE and UPDATE are arbitrarily differentiable functions (i.e., neural networks), and mN(u) is the “message,” which is aggregated from u's graph neighborhood N(u). We employ superscripts to identify the embeddings and functions at various message-passing iterations. The AGGREGATE function receives as input the set of embeddings of the nodes in the u's graph neighborhood N (u) at each iteration k of the GNN and generates a message. \({m}_{N(u)}^{k}\) . Based on this aggregated neighborhood information. The update function first UPDATES the message and then combines the message. \({m}_{N(u)}^{k}\) with the previous message \({h}_{u}^{(k-1)}\) of node, u to generate the updated message \({h}_{u}^{k}\) .

Description of GNN taxonomy

We can see from Fig. 8 below shows that we have divided our GNN taxonomy into 3 parts [ 30 ].

Graph Neural Network Taxonomy

1. Graph Structures 2. Graph Types 3. Graph Learning Tasks

Graph structure

The two scenarios shown in Fig. 9 typically present are structural and non-structural. Applications involving molecular and physical systems, knowledge graphs, and other objects explicitly state the graph structure in structural contexts.

Graph Structure

Graphs are implicit in non-structural situations. As a result, we must first construct the graph from the current task. For text, we must build a fully connected “a word” graph and a scene graph for images.

Graph types

There may be more information about nodes and links in complex graph types. Graphs are typically divided into 5 categories, as shown in Fig.  10 .

Types of Graphs

Directed/undirected graphs

A directed graph is characterized by edges with a specific direction, indicating the flow from one node to another. Conversely, in an undirected graph, the edges lack a designated direction, allowing nodes to interact bidirectionally. As illustrated in Fig. 11 (left side), the directed graph exhibits directed edges, while in Fig. 11 (right side), the undirected graph conspicuously lacks directional edges. In undirected graphs, it's important to note that each edge can be considered to comprise two directed edges, allowing for mutual interaction between connected nodes.

Directed/Undirected Graph

Static/dynamic graphs

The term “dynamic graph” pertains to a graph in which the properties or structure of the graph change with time. In dynamic graphs shown in Fig. 12 , it is essential to account for the temporal dimension appropriately. These dynamic graphs represent time-dependent events, such as the addition and removal of nodes and edges, typically presented as an ordered sequence or an asynchronous stream.

A noteworthy example of a dynamic graph can be observed in social networks like Twitter. In such networks, a new node is created each time a new user joins, and when a user follows another individual, a following edge is established. Furthermore, when users update their profiles, the respective nodes are also modified, reflecting the evolving nature of the graph. It's worth noting that different deep-learning libraries handle graph dynamics differently. TensorFlow, for instance, employs a static graph, while PyTorch utilizes a dynamic graph.

Static/Dynamic Graph

Homogeneous/heterogeneous graphs

Homogeneous graphs have only one type of node and one type of edge shown in Fig. 13 (Left). A homogeneous graph is one with the same type of nodes and edges, such as an online social network with friendship as edges and nodes representing people. In homogeneous networks, nodes and edges have the same types.

Heterogeneous graphs shown in Fig. 13 (Right) , however, have two or more different kinds of nodes and edges. A heterogeneous network is an online social network with various edges between nodes of the ‘person’ type, such as ‘friendship’ and ‘co-worker.’ Nodes and edges in heterogeneous graphs come in several varieties. Types of nodes and edges play critical functions in heterogeneous networks that require further consideration.

Homogeneous (Left), Heterogeneous (Right) Graph

Knowledge graphs

An array of triples in the form of (h, r, t) or (s, r, o) can be represented as a Knowledge Graph (KG), which is a network of entity nodes and relationship edges, with each triple (h, r, t) representing a single entity node. The relationship between an entity’s head (h) and tail (t) is denoted by the r. Knowledge Graph can be considered a heterogeneous graph from this perspective. The Knowledge Graph visually depicts several real-world objects and their relationships [ 32 ]. It can be used for many new aspects, including information retrieval, knowledge-guided innovation, and answering questions [ 30 ]. Entities are objects or things that exist in the real world, including individuals, organizations, places, music tracks, movies, and people. Each relation type describes a particular relationship between various elements similarly. We can see from Fig. 14 the Knowledge graph for Mr. Sundar Pichai.

Knowledge graph

Transductive/inductive graphs

In a transductive scenario shown in Fig. 15 (up), the entire graph is input, the label of the valid data is hidden, and finally, the label for the correct data is predicted. However, with an inductive graph shown in Fig. 15 (down), we also input the entire graph (but only sample to batch), mask the valid data’s label, and forecast the valuable data’s label. The model must forecast the labels of the given unlabeled nodes in a transductive context. In the inductive situation, it is possible to infer new unlabeled nodes from the same distribution.

Transductive/Inductive Graphs

Transductive Graph:

In the transductive approach, the entire graph is provided as input.

This method involves concealing the labels of the valid data.

The primary objective is to predict the labels for the valid data.

Inductive Graph:

The inductive approach still uses the complete graph, but only a sample within a batch is considered.

A crucial step in this process is masking the labels of the valid data.

The key aim here is to make predictions for the labels of the valid data.

Graph learning tasks

We perform three tasks with graphs: node classification, link prediction, and Graph Classification shown in Fig. 16 .

Node Level Prediction (e.g., social network) (LEFT), Edge Level Prediction (e.g., Next YouTube Video?) (MIDDLE), Graph Level Prediction (e.g., molecule) (Right)

Node-level task

Node-level tasks are primarily concerned with determining the identity or function of each node within a graph. The core objective of a node-level task is to predict specific properties associated with individual nodes. For example, a node-level task in social networks could involve predicting which social group a new member is likely to join based on their connections and the characteristics of their friends' memberships. Node-level tasks are typically used when working with unlabeled data, such as identifying whether a particular individual is a smoker.

Edge-level task (link prediction)

Edge-level tasks revolve around analyzing relationships between pairs of nodes in a graph. An illustrative application of an edge-level task is assessing the compatibility or likelihood of a connection between two entities, as seen in matchmaking or dating apps. Another instance of an edge-level task is evident when using platforms like Netflix, where the task involves predicting the following video to be recommended based on viewing history and user preferences.

Graph-level

In graph-level tasks, the objective is to make predictions about a characteristic or property that encompasses the entire graph. For example, using a graph-based representation, one might aim to predict attributes like the olfactory quality of a molecule or its potential to bind with a disease-associated receptor. The essence of a graph-level task is to provide predictions that pertain to the graph as a whole. For instance, when assessing a newly synthesized chemical compound, a graph-level task might seek to determine whether the molecule has the potential to be an effective drug. The summary of all three learning tasks are shown in Fig. 17 .

Graph Learning Tasks Summary

GNN models and comparative analysis of GNN models

Graph Neural Network (GNN) models represent a category of neural networks specially crafted to process data organized in graph structures. They've garnered substantial acclaim across various domains, primarily due to their exceptional capability to grasp intricate relationships and patterns within graph data. As illustrated in Fig.  18 , we've outlined three distinct GNN models. A comprehensive description of these GNN models, specifically Graph Convolutional Networks (GCN), Graph Attention Networks (GAT/GAN), and GraphSAGE models can be found in the reference [ 33 ]. In Sect. " GNN models ", we delve into these GNN models' intricacies; in " Comparative Study of GNN Models " section, we provide an in-depth analysis that explores their theoretical and practical aspects.

Graph convolution neural network (GCN)

GCN is one of the basic graph neural network variants. Thomas Kipf and Max Welling developed GCN networks. Convolution layers in Convolutional Neural Networks are essentially the same process as 'convolution' in GCNs. The input neurons are multiplied by weights called filters or kernels. The filters act as a sliding window across the image, allowing CNN to learn information from nearby cells. Weight sharing uses the same filter within the same layer throughout the image; when CNN is used to identify photos of cats vs. non-cats, the same filter is employed in the same layer to detect the cat's nose and ears. Throughout the image, the same weight (or kernel or filter in CNNs) is applied [ 33 ]. GCNs were first introduced in “Spectral Networks and Deep Locally Connected Networks on Graphs” [ 34 ].

GCNs, which learn features by analyzing neighboring nodes, carry out similar behaviors. The primary difference between CNNs and GNNs is that CNNs are made to operate on regular (Euclidean) ordered data. GNNs, on the other hand, are a generalized version of CNNs with different numbers of node connections and unordered nodes (irregular on non-Euclidean structured data). GCNs have been applied to solve many problems, for example, image classification [ 35 ], traffic forecasting [ 36 ], recommendation systems [ 17 ], scene graph generation [ 37 ], and visual question answering [ 38 ].

GCNs are particularly well-suited for tasks that involve data represented as graphs, such as social networks, citation networks, recommendation systems, and more. These networks are an extension of traditional CNNs, widely used for tasks involving grid-like data, such as images. The key idea behind GCNs is to perform convolution operations on the graph data. This enables them to capture and propagate information through the nodes in a graph by considering both a node’s features and those of its neighboring nodes. GCNs typically consist of several layers, each performing convolution and aggregation steps to refine the node representations in the graph. By applying these layers iteratively, GCNs can capture complex patterns and dependencies within the graph data.

Working of graph convolutional network

A Graph Convolutional Network (GCN) is a type of neural network architecture designed for processing and analyzing graph-structured data. GCNs work by aggregating and propagating information through the nodes in a graph. GCN works with the following steps shown in Fig.  19 :

Initialization:

Working of GCN

Each node in the graph is associated with a feature vector. Depending on the application, these feature vectors can represent various attributes or characteristics of the nodes. For example, in a social network, each node might represent a user, and the features could include user profile information.

Convolution Operation:

The core of a GCN is the convolution operation, which is adapted from convolutional neural networks (CNNs). It aims to aggregate information from neighboring nodes. This is done by taking a weighted sum of the feature vectors of neighboring nodes. The graph's adjacency matrix determines the weights. The resulting aggregated information is a new feature vector for each node.

Weighted Aggregation:

The graph's adjacency matrix, typically after normalization, provides weights for the aggregation process. In this context, for a given node, the features of its neighboring nodes are scaled by the corresponding values within the adjacency matrix, and the outcomes are then accumulated. A precise mathematical elucidation of this aggregation step is described in " Equation of GCN " section.

Activation function and learning weights:

The aggregated features are typically passed through an activation function (e.g., ReLU) to introduce non-linearity. The weight matrix W used in the aggregation step is learned during training. This learning process allows the GCN to adapt to the specific graph and task it is designed for.

Stacking Layers:

GCNs are often used in multiple layers. This allows the network to capture more complex relationships and higher-level features in the graph. The output of one GCN layer becomes the input for the next, and this process is repeated for a predefined number of layers.

Task-Specific Output:

The final output of the GCN can be used for various graph-based tasks, such as node classification, link prediction, or graph classification, depending on the specific application.

Equation of GCN

The Graph Convolutional Network (GCN) is based on a message-passing mechanism that can be described using mathematical equations. The core equation of a superficial, first-order GCN layer can be expressed as follows: For a graph with N nodes, let's define the following terms:

Equation  5.1 depicts a GCN layer's design. The normalized graph adjacency matrix A' and the nodes feature matrix F serve as the layer's inputs. The bias vector b and the weight matrix W are trainable parameters for the layer.

When used with the design matrix, the normalized adjacency matrix effectively smoothes a node’s feature vector based on the feature vectors of its close graph neighbors. This matrix captures the graph structure. A’ is normalized to make each neighboring node’s contribution proportional to the network's connectivity.

The layer definition is finished by applying A'FW + b to an element-wise non-linear function, such as ReLU. The downstream node classification task requires deep neural architectures to learn a complicated hierarchy of node attributes. This layer's output matrix Z can be routed into another GCN layer or any other neural network layer to do this.

Summary of graph convolution neural network (GCN) is shown in Table 2 .

Graph attention network (gat/gan).

Graph Attention Network (GAT/GAN) is a new neural network that works with graph-structured data. It uses masked self-attentional layers to address the shortcomings of past methods that depended on graph convolutions or their approximations. By stacking layers, the process makes it possible (implicitly) to assign various nodes in a neighborhood different weights, allowing nodes to focus on the characteristics of their neighborhoods without having to perform an expensive matrix operation (like inversion) or rely on prior knowledge of the graph's structure. GAT concurrently tackles numerous significant limitations of spectral-based graph neural networks, making the model suitable for both inductive and transductive applications.

Working of GAT

The Graph Attention Network (GAT) is a neural network architecture designed for processing and analyzing graph-structured data shown in Fig. 20 . GATs are a variation of Graph Convolutional Networks (GCNs) that incorporate the concept of attention mechanisms. GAT/GAN works with the following steps shown in Fig.  21 .

How attention Coefficients updates

As with other graph-based models, GAT starts with nodes in the graph, each associated with a feature vector. These features can represent various characteristics of the nodes.

Self-Attention Mechanism and Attention Computation:

GAT introduces an attention mechanism similar to what is used in sequence-to-sequence models in natural language processing. The attention mechanism allows each node to focus on different neighbors when aggregating information. It assigns different attention coefficients to the neighboring nodes, making the process more flexible. For each node in the graph, GAT computes attention scores for its neighboring nodes. These attention scores are based on the features of the central node and its neighbors. The attention scores are calculated using a weighted sum of the features of the central node and its neighbors.

The attention scores determine how much each neighbor’s feature contributes to the aggregation for the central node. This weighted aggregation is carried out for all neighboring nodes, resulting in a new feature vector for the central node.

Multiple Attention Heads and Output Combination:

GAT often employs multiple attention heads in parallel. Each attention head computes its attention scores and aggregation results. These multiple attention heads capture different aspects of the relationships in the graph. The outputs from the multiple attention heads are combined, typically by concatenation or averaging, to create a final feature vector for each node.

Learning Weights and Stacking Layers:

Similar to GCNs, GATs learn weight parameters during training. These weights are learned to optimize the attention mechanisms and adapt to the specific graph and task. GATs can be used in multiple layers to capture higher-level features and complex relationships in the graph. The output of one GAT layer becomes the input for the next layer.

The learning weights capture the importance of node relationships and contribute to information aggregation during the neighborhood aggregation process. The learning process in GNNs also relies on backpropagation and optimization algorithms. The stacking of GNN layers enables the model to capture higher-level abstractions and dependencies in the graph. Each layer refines the node representations based on information from the previous layer.

The final output of the GAT can be used for various graph-based tasks, such as node classification, link prediction, or graph classification, depending on the application.

Equation for GAT

GAT’s main distinctive feature is gathering data from the one-hop neighborhood [ 30 ]. A graph convolution operation in GCN produces the normalized sum of node properties of neighbors. Equation  5.2 shows the Graph attention network, which \({h}_{i}^{(l+1)}\) defines the current node output, \(\sigma\) denotes the non-linearity ReLU function, \(j\varepsilon N\left(i\right)\) one hop neighbor, \({\complement }_{i,j}\) normalized vector, \({W}^{\left(l\right)}\) weight matrix, and \({h}_{j}^{(l)}\) denotes the previous node.

Why is GAT better than GCN?

We learned from the Graph Convolutional Network (GCN) that integrating local graph structure and node-level features results in good node classification performance. The way GCN aggregates messages, on the other hand, is structure-dependent, which may limit its use.

How attention coefficients update: the attention layer has 4 parts: [ 47 ]

A linear transformation: A shared linear transformation is applied to each node in the following Equation.

where h is a set of node features. W is the weight matrix. Z is the output layer node.

Attention Coefficients: In the GAT paradigm, it is crucial because every node can now attend to every other node, discarding any structural information. The pair-wise un-normalized attention score between two neighbors is computed in the next step. It combines the 'z' embeddings of the two nodes. Where || stands for concatenation, a learnable weight vector a(l) is put through a dot product, and a LeakyReLU is used [ 1 ]. Contrary to the dot-product attention utilized in the Transformer model, this kind of attention is called additive attention. The nodes are subsequently subjected to self-attention.

Softmax: We utilize the softmax function to normalize the coefficients over all j values, improving their comparability across nodes.

Aggregation: This process is comparable to GCN. The neighborhood embeddings are combined and scaled based on the attention scores.

Summary of graph attention network (GAT) is shown in Table 3 .

GraphSAGE represents a tangible realization of an inductive learning framework shown in Fig. 22 . It exclusively considers training samples linked to the training set's edges during training. This process consists of two main steps: “Sampling” and “Aggregation.” Subsequently, the node representation vector is paired with the vector from the aggregated model and passed through a fully connected layer with a non-linear activation function. It's important to note that each network layer shares a standard aggregator and weight matrix. Thus, the consideration should be on the number of layers or weight matrices rather than the number of aggregators. Finally, a normalization step is applied to the layer's output.

Two major steps:

Sample It describes how to sample a large number of neighbors.

Aggregator refers to obtaining the neighbor node embedding and then determining how to collect these embeddings and change your embedding information.

Working of Graph SAGE Method

Working of graphSAGE model:

First, initializes the eigenvectors of all nodes in the input graph

For each node, get its sampled neighbor nodes

The aggregation function is used to aggregate the information of neighbor nodes

And combined with embedding, Update the same by a non-linear transformation embedding Express.

Types of aggregators

In the GraphSAGE method, 4 types of Aggregators are used.

Simple neighborhood aggregator:

Mean aggregator

LSTM Aggregator: Applies LSTM to a random permutation of neighbors.

Pooling Aggregator: It applies a symmetric vector function and converts adjacent vectors.

Equation of graphSAGE

W k , B k : is learnable weight matrices.

\({W}_{k}{B}_{k}=\) is learnable wight matrices.

\({h}_{v}^{0}= {x}_{v}:initial 0-\) the layer embeddings are equal to node features.

\({h}_{u}^{k-1}=\) Generalized Aggregation.

\({z}_{v }= {h}_{v}^{k}n\) : embedding after k layers of neighborhood aggregation.

\(\sigma\) – non linearity (ReLU).

Summary of graphSAGE is shown in Table 4 .

Comparative study of gnn models, comparison based on practical implementation of gnn models.

Table 5 describes the dataset statistics for different datasets used in literature for graph type of input. The datasets are CORA, Citeseer, and Pubmed. These statistics provide information about the kind of dataset, the number of nodes and edges, the number of classes, the number of features, and the label rate for each dataset. These details are essential for understanding the characteristics and scale of the datasets used in the context of citation networks. Comparison of the GNN model with equation in shown in Fig.  23 .

Equations of GNN Models

Table 6 shows the performance results of different Graph Neural Network (GNN) models on various datasets. Table 6 provides accuracy scores for other GNN models on different datasets. Additionally, the time taken for some models to compute results is indicated in seconds. This information is crucial for evaluating the performance of these models on specific datasets.

Comparison based on theoretical concepts of GNN models are described in Table 7 .

Graph neural network applications, graph construction.

Graph Neural Networks (GNNs) have a wide range of applications spanning diverse domains, which encompass modern recommender systems, computer vision, natural language processing, program analysis, software mining, bioinformatics, anomaly detection, and urban intelligence, among others. The fundamental prerequisite for GNN utilization is the transformation or representation of input data into a graph-like structure. In the realm of graph representation learning, GNNs excel in acquiring essential node or graph embeddings that serve as a crucial foundation for subsequent tasks [ 61 ].

The construction of a graph involves a two-fold process:

Graph creation and

Learning about graph representations

Graph Creation: The generation of graphs is essential for depicting the intricate relationships embedded within diverse incoming data. With the varied nature of input data, various applications adopt techniques to create meaningful graphs. This process is indispensable for effectively communicating the structural nuances of the data, ensuring the nodes and edges convey their semantic significance, particularly tailored to the specific task at hand.

Learning about graph representations: The subsequent phase involves utilizing the graph expression acquired from the input data. In GNN-based Learning for graph representations, some studies employ well-established GNN models like GraphSAGE, GCN, GAT, and GGNN, which offer versatility for various application tasks. However, when faced with specific tasks, it may be necessary to customize the GNN architecture to address particular challenges more effectively.

The different application which is considered a graph

Molecular Graphs: Atoms and electrons serve as the basic building blocks of matter and molecules, organized in three-dimensional structures. While all particles interact, we primarily acknowledge a covalent connection between two stable atoms when they are sufficiently spaced apart. Various atom-to-atom bond configurations exist, including single and double bonds. This three-dimensional arrangement is conveniently and commonly represented as a graph, with atoms representing nodes and covalent bonds representing edges [ 62 ].

Graphs of social networks: These networks are helpful research tools for identifying trends in the collective behavior of individuals, groups, and organizations. We may create a graph that represents groupings of people by visualizing individuals as nodes and their connections as edges [ 63 ].

Citation networks as graphs: When they publish papers, scientists regularly reference the work of other scientists. Each manuscript can be visualized as a node in a graph of these citation networks, with each directed edge denoting a citation from one publication to another. Additionally, we can include details about each document in each node, such as an abstract's word embedding [ 64 ].

Within computer vision: We may want to tag certain things in visual scenes. Then, we can construct graphs by treating these things as nodes and their connections as edges.

GNNs are used to model data as graphs, allowing for the capture of complex relationships and dependencies that traditional machine learning models may struggle to represent. This makes GNNs a valuable tool for tasks where data has an inherent graph structure or where modeling relationships is crucial for accurate predictions and analysis.

Graph neural networks (GNNs) applications in different fields

Nlp (natural language processing).

Document Classification: GNNs can be used to model the relationships between words or sentences in documents, allowing for improved document classification and information retrieval.

Text Generation: GNNs can assist in generating coherent and contextually relevant text by capturing dependencies between words or phrases.

Question Answering: GNNs can help in question-answering tasks by representing the relationships between question words and candidate answers within a knowledge graph.

Sentiment Analysis: GNNs can capture contextual information and sentiment dependencies in text, improving sentiment analysis tasks.

Computer vision

Image Segmentation: GNNs can be employed for pixel-level image segmentation tasks by modeling relationships between adjacent pixels as a graph.

Object Detection: GNNs can assist in object detection by capturing contextual information and relationships between objects in images.

Scene Understanding: GNNs are used for understanding complex scenes and modeling spatial relationships between objects in an image.

Bioinformatics

Protein-Protein Interaction Prediction: GNNs can be applied to predict interactions between proteins in biological networks, aiding in drug discovery and understanding disease mechanisms.

Genomic Sequence Analysis: GNNs can model relationships between genes or genetic sequences, helping in gene expression prediction and sequence classification tasks.

Drug Discovery: GNNs can be used for drug-target interaction prediction and molecular property prediction, which is vital in pharmaceutical research.

Table 8 offers a concise overview of various research papers that utilize Graph Neural Networks (GNNs) in diverse domains, showcasing the applications and contributions of GNNs in each study.

Table 9 highlights various applications of GNNs in Natural Language Processing, Computer Vision, and Bioinformatics domains, showcasing how GNN models are adapted and used for specific tasks within each field.

Future directions of graph neural network

The contribution of the existing literature to GNN principles, models, datasets, applications, etc., was the main emphasis of this survey. In this section, several potential future study directions are suggested. Significant challenges have been noted, including unbalanced datasets, the effectiveness of current methods, text classification, etc. We have also looked at the remedies to address these problems. We have suggested future and advanced directions to address these difficulties regarding domain adaptation, data augmentation, and improved classification. Table 10 displays future directions.

Imbalanced Datasets—Limited labeled data, domain-dependent data, and imbalanced data are currently issues with available datasets. Transfer learning and domain adaptation are solutions to these issues.

Accuracy of Existing Systems/Models—can utilize deep learning models such as GCN, GAT, and GraphSAGE approaches to increase the efficiency and precision of current systems. Additionally, training models on sizable, domain-specific datasets can enhance performance.

Enhancing Text Classification: Text classification poses another significant challenge, which is effectively addressed by leveraging advanced deep learning methodologies like graph neural networks, contributing to the improvement of text classification accuracy and performance.

The above Table  10 describes the research gaps and future directions presented in the above literature. These research gaps and future directions highlight the challenges and proposed solutions in the field of text classification and structural analysis.

Table 11 provides an overview of different research papers, their publication years, the applications they address, the graph structures they use, the graph types, the graph tasks, and the specific Graph Neural Network (GNN) models utilized in each study.

Conclusions

Graph Neural Networks (GNNs) have witnessed rapid advancements in addressing the unique challenges presented by data structured as graphs, a domain where conventional deep learning techniques, originally designed for images and text, often struggle to provide meaningful insights. GNNs offer a powerful and intuitive approach that finds broad utility in applications relying on graph structures. This comprehensive survey on GNNs offers an in-depth analysis covering critical aspects such as GNN fundamentals, the interplay with convolutional neural networks, GNN message-passing mechanisms, diverse GNN models, practical use cases, and a forward-looking perspective. Our central focus is on elucidating the foundational characteristics of GNNs, a field teeming with contemporary applications that continually enhance our comprehension and utilization of this technology.

The continuous evolution of GNN-based research has underscored the growing need to address issues related to graph analysis, which we aptly refer to as the frontiers of GNNs. In our exploration, we delve into several crucial recent research domains within the realm of GNNs, encompassing areas like link prediction, graph generation, and graph categorization, among others.

Availability of data and materials

Not applicable.

Abbreviations

Graph Neural Network

Graph Convolution Network

Graph Attention Networks

Natural Language Processing

Convolution Neural Networks

Recurrent Neural Networks

Machine Learning

Deep Learning

Knowledge Graph

Pucci A, Gori M, Hagenbuchner M, Scarselli F, Tsoi AC. Investigation into the application of graph neural networks to large-scale recommender systems, infona.pl, no. 32, no 4, pp. 17–26, 2006.

Mahmud FB, Rayhan MM, Shuvo MH, Sadia I, Morol MK. A comparative analysis of Graph Neural Networks and commonly used machine learning algorithms on fake news detection, Proc. - 2022 7th Int. Conf. Data Sci. Mach. Learn. Appl. CDMA 2022, pp. 97–102, 2022.

Cui L, Seo H, Tabar M, Ma F, Wang S, Lee D, Deterrent: Knowledge Guided Graph Attention Network for Detecting Healthcare Misinformation, Proc. ACM SIGKDD Int. Conf. Knowl. Discov. Data Min., pp. 492–502, 2020.

Gori M, Monfardini G, Scarselli F, A new model for earning in raph domains, Proc. Int. Jt. Conf. Neural Networks, vol. 2, no. January 2005, pp. 729–734, 2005, https://doi.org/10.1109/IJCNN.2005.1555942 .

Scarselli F, Yong SL, Gori M, Hagenbuchner M, Tsoi AC, Maggini M. Graph neural networks for ranking web pages, Proc.—2005 IEEE/WIC/ACM Int. Web Intell. WI 2005, vol. 2005, no. January, pp. 666–672, 2005, doi: https://doi.org/10.1109/WI.2005.67 .

Gandhi S, Zyer AP, P3: Distributed deep graph learning at scale, Proc. 15th USENIX Symp. Oper. Syst. Des. Implementation, OSDI 2021, pp. 551–568, 2021.

Li C, Guo J, Zhang H. Pruning neighborhood graph for geodesic distance based semi-supervised classification, in 2007 International Conference on Computational Intelligence and Security (CIS 2007), 2007, pp. 428–432.

Zhang Z, Cui P, Pei J, Wang X, Zhu W, Eigen-gnn: A graph structure preserving plug-in for gnns, IEEE Trans. Knowl. Data Eng., 2021.

Nandedkar AV, Biswas PK. A granular reflex fuzzy min–max neural network for classification. IEEE Trans Neural Netw. 2009;20(7):1117–34.

Article   Google Scholar  

Chaturvedi DK, Premdayal SA, Chandiok A. Short-term load forecasting using soft computing techniques. Int’l J Commun Netw Syst Sci. 2010;3(03):273.

Google Scholar  

Hashem T, Kulik L, Zhang R. Privacy preserving group nearest neighbor queries, in Proceedings of the 13th International Conference on Extending Database Technology, 2010, pp. 489–500.

Sun Z et al. Knowledge graph alignment network with gated multi-hop neighborhood aggregation, in Proceedings of the AAAI Conference on Artificial Intelligence, 2020, vol. 34, no. 01, pp. 222–229.

Zhang M, Chen Y. Link prediction based on graph neural networks. Adv Neural Inf Process Syst. 31, 2018.

Stanimirović PS, Katsikis VN, Li S. Hybrid GNN-ZNN models for solving linear matrix equations. Neurocomputing. 2018;316:124–34.

Stanimirović PS, Petković MD. Gradient neural dynamics for solving matrix equations and their applications. Neurocomputing. 2018;306:200–12.

Zhang C, Song D, Huang C, Swami A, Chawla NV. Heterogeneous graph neural network, in Proceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining, 2019, pp. 793–803.

Fan W et al. Graph neural networks for social recommendation," in The world wide web conference, 2019, pp. 417–426.

Gui T et al. A lexicon-based graph neural network for Chinese NER," in Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), 2019, pp. 1040–1050.

Qasim SR, Mahmood H, Shafait F. Rethinking table recognition using graph neural networks, in 2019 International Conference on Document Analysis and Recognition (ICDAR), 2019, pp. 142–147

You J, Ying R, Leskovec J. Position-aware graph neural networks, in International conference on machine learning, 2019, pp. 7134–7143.

Cao D, et al. Spectral temporal graph neural network for multivariate time-series forecasting. Adv Neural Inf Process Syst. 2020;33:17766–78.

Xhonneux LP, Qu M, Tang J. Continuous graph neural networks. In International Conference on Machine Learning, 2020, pp. 10432–10441.

Zhou K, Huang X, Li Y, Zha D, Chen R, Hu X. Towards deeper graph neural networks with differentiable group normalization. Adv Neural Inf Process Syst. 2020;33:4917–28.

Gu F, Chang H, Zhu W, Sojoudi S, El Ghaoui L. Implicit graph neural networks. Adv Neural Inf Process Syst. 2020;33:11984–95.

Liu Y, Guan R, Giunchiglia F, Liang Y, Feng X. Deep attention diffusion graph neural networks for text classification. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021, pp. 8142–8152.

Gasteiger J, Becker F, Günnemann S. Gemnet: universal directional graph neural networks for molecules. Adv Neural Inf Process Syst. 2021;34:6790–802.

Yao D et al. Deep hybrid: multi-graph neural network collaboration for hyperspectral image classification. Def. Technol. 2022.

Li Y, et al. Research on multi-port ship traffic prediction method based on spatiotemporal graph neural networks. J Mar Sci Eng. 2023;11(7):1379.

Djenouri Y, Belhadi A, Srivastava G, Lin JC-W. Hybrid graph convolution neural network and branch-and-bound optimization for traffic flow forecasting. Futur Gener Comput Syst. 2023;139:100–8.

Zhou J, et al. Graph neural networks: a review of methods and applications. AI Open. 2020;1(January):57–81. https://doi.org/10.1016/j.aiopen.2021.01.001 .

Rong Y, Huang W, Xu T, Huang J. Dropedge: Towards deep graph convolutional networks on node classification. arXiv preprint arXiv:1907.10903. 2019.

Abu-Salih B, Al-Qurishi M, Alweshah M, Al-Smadi M, Alfayez R, Saadeh H. Healthcare knowledge graph construction: a systematic review of the state-of-the-art, open issues, and opportunities. J Big Data. 2023;10(1):81.

Kipf TN, Welling M. Semi-supervised classification with graph convolutional networks. arXiv Prepr. arXiv1609.02907, 2016.

Berg RV, Kipf TN, Welling M. Graph Convolutional Matrix Completion. 2017, http://arxiv.org/abs/1706.02263

Monti F, Boscaini D, Masci J, Rodola E, Svoboda J, Bronstein MM. Geometric deep learning on graphs and manifolds using mixture model cnns. InProceedings of the IEEE conference on computer vision and pattern recognition 2017 (pp. 5115-5124).

Cui Z, Henrickson K, Ke R, Wang Y. Traffic graph convolutional recurrent neural network: a deep learning framework for network-scale traffic learning and forecasting. IEEE Trans Intell Transp Syst. 2020;21(11):4883–94. https://doi.org/10.1109/TITS.2019.2950416 .

Yang J, Lu J, Lee S, Batra D, Parikh D. Graph r-cnn for scene graph generation. InProceedings of the European conference on computer vision (ECCV) 2018 (pp. 670-685). https://doi.org/10.1007/978-3-030-01246-5_41 .

Teney D, Liu L, van Den Hengel A. Graph-structured representations for visual question answering. InProceedings of the IEEE conference on computer vision and pattern recognition 2017 (pp. 1-9). https://doi.org/10.1109/CVPR.2017.344 .

Yao L, Mao C, Luo Y. Graph convolutional networks for text classification. Proc AAAI Conf Artif Intell. 2019;33(01):7370–7.

De Cao N, Aziz W, Titov I. Question answering by reasoning across documents with graph convolutional networks. arXiv Prepr. arXiv1808.09920, 2018.

Gao H, Wang Z, Ji S. Large-scale learnable graph convolutional networks. in Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, 2018, pp. 1416–1424.

Hu F, Zhu Y, Wu S, Wang L, Tan T. Hierarchical graph convolutional networks for semi-supervised node classification. arXiv Prepr. arXiv1902.06667, 2019.

Lange O, Perez L. Traffic prediction with advanced graph neural networks. DeepMind Research Blog Post, https://deepmind.google/discover/blog/traffic-prediction-with-advanced-graph-neural-networks/ . 2020.

Duan C, Hu B, Liu W, Song J. Motion capture for sporting events based on graph convolutional neural networks and single target pose estimation algorithms. Appl Sci. 2023;13(13):7611.

Balcıoğlu YS, Sezen B, Çerasi CC, Huang SH. machine design automation model for metal production defect recognition with deep graph convolutional neural network. Electronics. 2023;12(4):825.

Baghbani A, Bouguila N, Patterson Z. Short-term passenger flow prediction using a bus network graph convolutional long short-term memory neural network model. Transp Res Rec. 2023;2677(2):1331–40.

Velickovic P, Cucurull G, Casanova A, Romero A, Lio P, Bengio Y. Graph attention networks. Stat. 2017;1050(20):10–48550.

Hamilton W, Ying Z, Leskovec J. Inductive representation learning on large graphs. Advances in neural information processing systems. 2017; 30.

Ye Y, Ji S. Sparse graph attention networks. IEEE Trans Knowl Data Eng. 2021;35(1):905–16.

MathSciNet   Google Scholar  

Chen Z et al. Graph neural network-based fault diagnosis: a review. arXiv Prepr. arXiv2111.08185, 2021.

Brody S, Alon U, Yahav E. How attentive are graph attention networks? arXiv Prepr. arXiv2105.14491, 2021.

Huang J, Shen H, Hou L, Cheng X. Signed graph attention networks," in International Conference on Artificial Neural Networks. 2019, pp. 566–577.

Seraj E, Wang Z, Paleja R, Sklar M, Patel A, Gombolay M. Heterogeneous graph attention networks for learning diverse communication. arXiv preprint arXiv: 2108.09568. 2021.

Zhang Y, Wang X, Shi C, Jiang X, Ye Y. Hyperbolic graph attention network. IEEE Transactions on Big Data. 2021;8(6):1690–701.

Yang X, Ma H, Wang M. Research on rumor detection based on a graph attention network with temporal features. Int J Data Warehous Min. 2023;19(2):1–17.

Lan W, et al. KGANCDA: predicting circRNA-disease associations based on knowledge graph attention network. Brief Bioinform. 2022;23(1):bbab494.

Xiao L, Wu X, Wang G, 2019, December. Social network analysis based on graph SAGE. In 2019 12th international symposium on computational intelligence and design (ISCID) (Vol. 2, pp. 196–199). IEEE.

Chang L, Branco P. Graph-based solutions with residuals for intrusion detection: The modified e-graphsage and e-resgat algorithms. arXiv preprint arXiv:2111.13597. 2021.

Oh J, Cho K, Bruna J. Advancing graphsage with a data-driven node sampling. arXiv preprint arXiv:1904.12935. 2019.

Kapoor M, Patra S, Subudhi BN, Jakhetiya V, Bansal A. Underwater Moving Object Detection Using an End-to-End Encoder-Decoder Architecture and GraphSage With Aggregator and Refactoring. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition 2023 (pp. 5635-5644).

Bhatti UA, Tang H, Wu G, Marjan S, Hussain A. Deep learning with graph convolutional networks: An overview and latest applications in computational intelligence. Int J Intell Syst. 2023;2023:1–28.

David L, Thakkar A, Mercado R, Engkvist O. Molecular representations in AI-driven drug discovery: a review and practical guide. J Cheminform. 2020;12(1):1–22.

Davies A, Ajmeri N. Realistic Synthetic Social Networks with Graph Neural Networks. arXiv preprint arXiv:2212.07843. 2022; 15.

Frank MR, Wang D, Cebrian M, Rahwan I. The evolution of citation graphs in artificial intelligence research. Nat Mach Intell. 2019;1(2):79–85.

Gao C, Wang X, He X, Li Y. Graph neural networks for recommender system. InProceedings of the Fifteenth ACM International Conference on Web Search and Data Mining 2022 (pp. 1623-1625).

Wu S, Sun F, Zhang W, Xie X, Cui B. Graph neural networks in recommender systems: a survey. ACM Comput Surv. 2022;55(5):1–37.

Wu L, Chen Y, Shen K, Guo X, Gao H, Li S, Pei J, Long B. Graph neural networks for natural language processing: a survey. Found Trends Mach Learn. 2023;16(2):119–328.

Wu L, Chen Y, Ji H, Liu B. Deep learning on graphs for natural language processing. InProceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval 2021 (pp. 2651-2653).

Liu X, Su Y, Xu B. The application of graph neural network in natural language processing and computer vision. In2021 3rd International Conference on Machine Learning, Big Data and Business Intelligence (MLBDBI) 2021 (pp. 708-714).

Harmon SHE, Faour DE, MacDonald NE. Mandatory immunization and vaccine injury support programs: a survey of 28 GNN countries. Vaccine. 2021;39(49):7153–7.

Yan W, Zhang Z, Zhang Q, Zhang G, Hua Q, Li Q. Deep data analysis-based agricultural products management for smart public healthcare. Front Public Health. 2022;10:847252.

Hamaguchi T, Oiwa H, Shimbo M, Matsumoto Y. Knowledge transfer for out-of-knowledge-base entities: a graph neural network approach. arXiv preprint arXiv:1706.05674. 2017.

Dai D, Zheng H, Luo F, Yang P, Chang B, Sui Z. Inductively representing out-of-knowledge-graph entities by optimal estimation under translational assumptions. arXiv preprint arXiv:2009.12765.

Pradhyumna P, Shreya GP. Graph neural network (GNN) in image and video understanding using deep learning for computer vision applications. In2021 Second International Conference on Electronics and Sustainable Communication Systems (ICESC) 2021 (pp. 1183-1189).

Shi W, Rajkumar R. Point-gnn: Graph neural network for 3d object detection in a point cloud. InProceedings of the IEEE/CVF conference on computer vision and pattern recognition 2020 (pp. 1711-1719).

Wu Y, Dai HN, Tang H. Graph neural networks for anomaly detection in industrial internet of things. IEEE Int Things J. 2021;9(12):9214–31.

Pitsik EN, et al. The topology of fMRI-based networks defines the performance of a graph neural network for the classification of patients with major depressive disorder. Chaos Solitons Fractals. 2023;167: 113041.

Liao W, Zeng B, Liu J, Wei P, Cheng X, Zhang W. Multi-level graph neural network for text sentiment analysis. Comput Electr Eng. 2021;92: 107096.

Kumar VS, Alemran A, Karras DA, Gupta SK, Dixit CK, Haralayya B. Natural Language Processing using Graph Neural Network for Text Classification. In2022 International Conference on Knowledge Engineering and Communication Systems (ICKES) 2022; (pp. 1-5).

Dara S, Srinivasulu CH, Babu CM, Ravuri A, Paruchuri T, Kilak AS, Vidyarthi A. Context-Aware auto-encoded graph neural model for dynamic question generation using NLP. ACM transactions on asian and low-resource language information processing. 2023.

Wu L, Cui P, Pei J, Zhao L, Guo X. Graph neural networks: foundation, frontiers and applications. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2022; (pp. 4840-4841).

Scarselli F, Gori M, Tsoi AC, Hagenbuchner M, Monfardini G. The graph neural network model. IEEE Trans neural networks. 2008;20(1):61–80.

Cao P, Zhu Z, Wang Z, Zhu Y, Niu Q. Applications of graph convolutional networks in computer vision. Neural Comput Appl. 2022;34(16):13387–405.

You R, Yao S, Mamitsuka H, Zhu S. DeepGraphGO: graph neural network for large-scale, multispecies protein function prediction. Bioinformatics. 2021;37(Supplement_1):i262-71.

Long Y, et al. Pre-training graph neural networks for link prediction in biomedical networks. Bioinformatics. 2022;38(8):2254–62.

Wu Y, Gao M, Zeng M, Zhang J, Li M. BridgeDPI: a novel graph neural network for predicting drug–protein interactions. Bioinformatics. 2022;38(9):2571–8.

Kang C, Zhang H, Liu Z, Huang S, Yin Y. LR-GNN: a graph neural network based on link representation for predicting molecular associations. Briefings Bioinf. 2022;23(1):bbab513.

Wei X, Huang H, Ma L, Yang Z, Xu L. Recurrent Graph Neural Networks for Text Classification. in 2020 IEEE 11th International Conference on Software Engineering and Service Science (ICSESS), 2020, pp. 91–97.

Schlichtkrull MS, De Cao N, Titov I. Interpreting graph neural networks for nlp with differentiable edge masking. arXiv Prepr. arXiv2010.00577, 2020.

Tu M, Huang J, He X, Zhou B. Graph sequential network for reasoning over sequences. arXiv Prepr. arXiv2004.02001, 2020.

Download references

Acknowledgements

I am grateful to all of those with whom I have had the pleasure to work during this research work. Each member has provided me extensive personal and professional guidance and taught me a great deal about scientific research and life in general.

This work was supported by the Research Support Fund (RSF) of Symbiosis International (Deemed University), Pune, India.

Author information

Authors and affiliations.

Symbiosis Institute of Technology Pune Campus, Symbiosis International (Deemed University) (SIU), Lavale, Pune, 412115, India

Bharti Khemani

Symbiosis Centre for Applied Artificial Intelligence (SCAAI), Symbiosis Institute of Technology Pune Campus, Symbiosis International (Deemed University) (SIU), Lavale, Pune, 412115, India

Shruti Patil & Ketan Kotecha

IEEE, Department of Computer Science and Engineering, Institute of Technology, Nirma University, Ahmedabad, India

Sudeep Tanwar

You can also search for this author in PubMed   Google Scholar

Contributions

Conceptualization, BK and SP; methodology, BK and SP; software, BK; validation, BK, SP, KK; formal analysis, BK; investigation, BK; resources, BK; data curation, BK and SP; writing—original draft preparation, BK; writing—review and editing, SP, KK, and ST; visualization, BK; supervision, SP; project administration, SP, ST; funding acquisition, KK.

Corresponding author

Correspondence to Shruti Patil .

Ethics declarations

Ethics approval and consent to participate.

Not applicable

Consent for publication

Competing interests.

The authors declare that they have no competing interests .

Additional information

Publisher's note.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

See Tables  12 and 13

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/ .

Reprints and permissions

About this article

Cite this article.

Khemani, B., Patil, S., Kotecha, K. et al. A review of graph neural networks: concepts, architectures, techniques, challenges, datasets, applications, and future directions. J Big Data 11 , 18 (2024). https://doi.org/10.1186/s40537-023-00876-4

Download citation

Received : 28 June 2023

Accepted : 27 December 2023

Published : 16 January 2024

DOI : https://doi.org/10.1186/s40537-023-00876-4

Share this article

Anyone you share the following link with will be able to read this content:

Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative

• Graph Neural Network (GNN)
• Graph Convolution Network (GCN)
• Graph Attention Networks (GAT)
• Message Passing Mechanism
• Natural Language Processing (NLP)

• Affiliate Program

• UNITED STATES
• 台灣 (TAIWAN)
• TÜRKIYE (TURKEY)
• Academic Editing Services
• - Research Paper
• - Journal Manuscript
• - Dissertation
• - College & University Assignments
• Admissions Editing Services
• - Application Essay
• - Personal Statement
• - Recommendation Letter
• - Cover Letter
• - CV/Resume
• Business Editing Services
• - Business Documents
• - Report & Brochure
• - Website & Blog
• Writer Editing Services
• - Script & Screenplay
• Our Editors
• Client Reviews
• Editing & Proofreading Prices
• Wordvice Points
• Partner Discount
• Plagiarism Checker
• APA Citation Generator
• MLA Citation Generator
• Chicago Citation Generator
• Vancouver Citation Generator
• - APA Style
• - MLA Style
• - Chicago Style
• - Vancouver Style
• Writing & Editing Guide
• Academic Resources
• Admissions Resources

How to Use Tables & Graphs in a Research Paper

It might not seem very relevant to the story and outcome of your study, but how you visually present your experimental or statistical results can play an important role during the review and publication process of your article. A presentation that is in line with the overall logical flow of your story helps you guide the reader effectively from your introduction to your conclusion. 

If your results (and the way you organize and present them) don’t follow the story you outlined in the beginning, then you might confuse the reader and they might end up doubting the validity of your research, which can increase the chance of your manuscript being rejected at an early stage. This article illustrates the options you have when organizing and writing your results and will help you make the best choice for presenting your study data in a research paper.

Why does data visualization matter?

Your data and the results of your analysis are the core of your study. Of course, you need to put your findings and what you think your findings mean into words in the text of your article. But you also need to present the same information visually, in the results section of your manuscript, so that the reader can follow and verify that they agree with your observations and conclusions. 

The way you visualize your data can either help the reader to comprehend quickly and identify the patterns you describe and the predictions you make, or it can leave them wondering what you are trying to say or whether your claims are supported by evidence. Different types of data therefore need to be presented in different ways, and whatever way you choose needs to be in line with your story. 

Another thing to keep in mind is that many journals have specific rules or limitations (e.g., how many tables and graphs you are allowed to include, what kind of data needs to go on what kind of graph) and specific instructions on how to generate and format data tables and graphs (e.g., maximum number of subpanels, length and detail level of tables). In the following, we will go into the main points that you need to consider when organizing your data and writing your result section .

Table of Contents:

Types of data , when to use data tables .

• When to Use Data Graphs 

Common Types of Graphs in Research Papers 

Journal guidelines: what to consider before submission.

Depending on the aim of your research and the methods and procedures you use, your data can be quantitative or qualitative. Quantitative data, whether objective (e.g., size measurements) or subjective (e.g., rating one’s own happiness on a scale), is what is usually collected in experimental research. Quantitative data are expressed in numbers and analyzed with the most common statistical methods. Qualitative data, on the other hand, can consist of case studies or historical documents, or it can be collected through surveys and interviews. Qualitative data are expressed in words and needs to be categorized and interpreted to yield meaningful outcomes. 

Quantitative data example: Height differences between two groups of participants Qualitative data example: Subjective feedback on the food quality in the work cafeteria

Depending on what kind of data you have collected and what story you want to tell with it, you have to find the best way of organizing and visualizing your results.

When you want to show the reader in detail how your independent and dependent variables interact, then a table (with data arranged in columns and rows) is your best choice. In a table, readers can look up exact values, compare those values between pairs or groups of related measurements (e.g., growth rates or outcomes of a medical procedure over several years), look at ranges and intervals, and select specific factors to search for patterns. 

Tables are not restrained to a specific type of data or measurement. Since tables really need to be read, they activate the verbal system. This requires focus and some time (depending on how much data you are presenting), but it gives the reader the freedom to explore the data according to their own interest. Depending on your audience, this might be exactly what your readers want. If you explain and discuss all the variables that your table lists in detail in your manuscript text, then you definitely need to give the reader the chance to look at the details for themselves and follow your arguments. If your analysis only consists of simple t-tests to assess differences between two groups, you can report these results in the text (in this case: mean, standard deviation, t-statistic, and p-value), and do not necessarily need to include a table that simply states the same numbers again. If you did extensive analyses but focus on only part of that data (and clearly explain why, so that the reader does not think you forgot to talk about the rest), then a graph that illustrates and emphasizes the specific result or relationship that you consider the main point of your story might be a better choice.

When to Use Data Graphs

Graphs are a visual display of information and show the overall shape of your results rather than the details. If used correctly, a visual representation helps your (or your reader’s) brain to quickly understand large amounts of data and spot patterns, trends, and exceptions or outliers. Graphs also make it easier to illustrate relationships between entire data sets. This is why, when you analyze your results, you usually don’t just look at the numbers and the statistical values of your tests, but also at histograms, box plots, and distribution plots, to quickly get an overview of what is going on in your data.

Line graphs

When you want to illustrate a change over a continuous range or time, a line graph is your best choice. Changes in different groups or samples over the same range or time can be shown by lines of different colors or with different symbols.

Example: Let’s collapse across the different food types and look at the growth of our four fish species over time.

You should use a bar graph when your data is not continuous but divided into categories that are not necessarily connected, such as different samples, methods, or setups. In our example, the different fish types or the different types of food are such non-continuous categories.

Example: Let’s collapse across the food types again and also across time, and only compare the overall weight increase of our four fish types at the end of the feeding period.

Scatter plots

Scatter plots can be used to illustrate the relationship between two variables — but note that both have to be continuous. The following example displays “fish length” as an additional variable–none of the variables in our table above (fish type, fish food, time) are continuous, and they can therefore not be used for this kind of graph. 

As you see, these example graphs all contain less data than the table above, but they lead the reader to exactly the key point of your results or the finding you want to emphasize. If you let your readers search for these observations in a big table full of details that are not necessarily relevant to the claims you want to make, you can create unnecessary confusion. Most journals allow you to provide bigger datasets as supplementary information, and some even require you to upload all your raw data at submission. When you write up your manuscript, however, matching the data presentation to the storyline is more important than throwing everything you have at the reader. 

Don’t forget that every graph needs to have clear x and y axis labels , a title that summarizes what is shown above the figure, and a descriptive legend/caption below. Since your caption needs to stand alone and the reader needs to be able to understand it without looking at the text, you need to explain what you measured/tested and spell out all labels and abbreviations you use in any of your graphs once more in the caption (even if you think the reader “should” remember everything by now, make it easy for them and guide them through your results once more). Have a look at this article if you need help on how to write strong and effective figure legends .

Even if you have thought about the data you have, the story you want to tell, and how to guide the reader most effectively through your results, you need to check whether the journal you plan to submit to has specific guidelines and limitations when it comes to tables and graphs. Some journals allow you to submit any tables and graphs initially (as long as tables are editable (for example in Word format, not an image) and graphs of high enough resolution. 

Some others, however, have very specific instructions even at the submission stage, and almost all journals will ask you to follow their formatting guidelines once your manuscript is accepted. The closer your figures are already to those guidelines, the faster your article can be published. This PLOS One Figure Preparation Checklist is a good example of how extensive these instructions can be – don’t wait until the last minute to realize that you have to completely reorganize your results because your target journal does not accept tables above a certain length or graphs with more than 4 panels per figure. 

Some things you should always pay attention to (and look at already published articles in the same journal if you are unsure or if the author instructions seem confusing) are the following:

• How many tables and graphs are you allowed to include?
• What file formats are you allowed to submit?
• Are there specific rules on resolution/dimension/file size?
• Should your figure files be uploaded separately or placed into the text?
• If figures are uploaded separately, do the files have to be named in a specific way?
• Are there rules on what fonts to use or to avoid and how to label subpanels?
• Are you allowed to use color? If not, make sure your data sets are distinguishable.

If you are dealing with digital image data, then it might also be a good idea to familiarize yourself with the difference between “adjusting” for clarity and visibility and image manipulation, which constitutes scientific misconduct .  And to fully prepare your research paper for publication before submitting it, be sure to receive proofreading services , including journal manuscript editing and research paper editing , from Wordvice’s professional academic editors .

Connected Papers: a visual tool for academic research

Anne-Laure Le Cunff

I’m obsessed with thinking in maps : discovering and creating connections between ideas, adding nodes to a knowledge graph, finding patterns across distant areas of knowledge. However, the traditional way of exploring connections between research papers is fairly tedious: read the paper, scan the references, search for any relevant title, rinse and repeat. Connected Papers aims to shake things up.

Connected Papers is a tool for thought to help researchers and applied scientists find and explore papers relevant to their field of work in a visual way. You enter an origin paper, and they generate a graph. To achieve this, they analyse about 50,000 research papers, and select the ones with the strongest connections to the origin paper.

Created by Alex Tarnavsky, Eitan Eddie Smolyansky, and Itay Knaan Harpaz from Israel, Connected Papers started as a weekend side project. But when the three friends realised how useful it was in their own research, and how their friends and colleagues kept on asking to use it, they decided to release the tool for the public.

Some of the benefits of Connected Papers include:

• Getting a visual overview of a field of research. You will be able to see at a glance which papers are most popular in the field, as well as the various dynamics between areas of studies.
• Making sure you haven’t missed a key paper. This is especially useful in fields that constantly produce a large volume of new papers.
• Exploring relevant papers in a bi-directional manner. Connected Papers lets you discover the most important prior and derivative work in your area of interest.

The tool is currently completely free, and the three co-founders keep on adding new features to make it even more useful. If you want to give it a try, follow these instructions.

1. Enter an origin paper

On the home page, enter one of the options to identify your origin paper. You can use a DOI, the paper’s title, or the paper’s URL from arXiv, PubMed, or Semantic Scholar. Then click on “Build a graph. For this tutorial, I used this paper , which can read more about here .

2. Read the graph

On the next page, you will be greeted by three panels. We’ll discuss the other panels later, but for now, let’s focus on the graph. Each node is a research paper related to the origin paper. Rather than a basic citation tree, the papers are arranged according to their similarity.

The size of a node represents the number of citations. The color of a node represents the publishing year—lighter is older. You will notice that highly similar papers have stronger connecting lines and tend to cluster together.

3. Explore the graph

You can scroll through papers in the left panel. Whenever you click on a paper there, it will be highlighted on the graph. You can also navigate the graph by clicking on specific nodes. Both options will update the right-side panel with more information about the selected paper.

Two buttons in the top left corner allow you to explore papers that are not included in the graph, but probably relevant to your topic of choice.

• Prior works. These are research papers that were most commonly cited by the papers included in the graph. It usually means that they are important seminal works for this field. Selecting a prior work will highlight all graph papers referencing it in the left-side panel, and selecting a graph paper will highlight all referenced prior work.
• Derivative works. These are research papers that cited many of the graph papers. It probably means they are either recent relevant works or surveys of the field. Similar to prior works, “selecting a derivative work will highlight all graph papers cited by it, and selecting a graph paper will highlight all derivative works citing it.”

If you find a paper particularly promising, you can click on “paper details” to open the link to the paper in a new window, or on “build a graph” to create a new graph based on this origin paper. Building the new graph can sometimes take a few seconds, but there will be a progress bar so you know how long to wait.

All of your graphs can be found in the top right corner of the tool, under “my graphs”.

Connected Papers is incredibly well designed, easy to use, and most importantly very helpful in exploring research paths of influence. I highly recommend giving it a try to build your mental atlas .

Update: Connected Papers is now supported on mobile browsers !

Join 80,000 mindful makers!

Maker Mind is a weekly newsletter with science-based insights on creativity, mindful productivity, better thinking and lifelong learning.

One email a week, no spam, ever. See our Privacy policy .

Don’t work more. Work mindfully.

Ness Labs provides content, coaching, courses and community to help makers put their minds at work. Apply evidence-based strategies to your daily life, discover the latest in neuroscience research, and connect with fellow mindful makers.

Ness Labs © 2022. All rights reserved .

Using Tables and Graphs for Reporting Data

Affiliation.

• 1 Professor and Head, Department of Pediatrics, Topiwala National Medical College and BYL Nair Ch. Hospital, Mumbai, Maharashtra.
• PMID: 27608693

For presenting the study results, authors need to choose from one of the three formats: text, tables and graphs. The underlying consideration of any author while making the choice is to ensure that readers find it easy to understand, to assist the author present data in a way that would catch the reader's eye, hold his interest and enhance his understanding. The choice should, therefore, be based on the advantages and limitations of each of these formats. Most of the observations are reported in textual format. Tables are useful when comparisons are to be shown. Graphs attract readers' attention better and the data they depict remains in the reader's memory. The type of graph used is dependent upon the nature of data that is to be shown. Every research paper has a lot to say through the data that has been collected and analyzed. However, this data needs to be presented in a logical, coherent and easy-to-understand manner. The way data is presented can influence the reviewers' and editors' decision. It can also make the difference between a paper being appreciated by the readers or being neglected or even trashed by them. Tables and Graphs are the two important tools that authors use to make the data more presentable and easy to read and understand.

© Journal of the Association of Physicians of India 2011.

• Computer Graphics*
• Research Design*
• Research Report*

How to Create Precise Graphs, Diagrams or Images in a Research Paper

According to the American psychologist Howard Gardner, human intelligence can be divided into seven categories: visual-spatial, bodily kinesthetic, musical, interpersonal, intrapersonal, linguistic, and logical-mathematical. This implies our intelligence strengths can be different in each (so-called) intelligence profile and that everybody can be intelligent in many different ways.

Gardner says these differences “challenge an educational system that assumes that everyone can learn the same materials in the same way and that a uniform, universal measure suffices to test student learning.” The truth is that we learn and understand things differently, and these differences affect the manner we read academic papers. A research paper is usually a combination of written and visual information. We can assume that those who have a predominant linguistic intelligence would focus on written information, whereas those with a visual-spatial intelligence would feel more comfortable focusing on graphs, diagrams, or images. How to combine both to achieve a paper that engages readers with different intelligence profiles at par?

The Perfect Combination

The first thing we must understand is that, no matter how much visual support they have, papers are written works. Filling pages with unnecessary images, graphs, diagrams or any other kind of visual material is never a good idea. Remember that you are writing a professional academic paper and, therefore, your capacity to discern which material is important. Once this is clear, it is time to discern which information is likely to be visually demonstrated .

Some main ideas would help you to decide when to use graphs. Choose only information that can be clearer if explained visually, and only if it is so important that you desire the reader to keep focus on it more than in other parts. Besides, this piece of information must be qualitatively or quantitatively measurable.

Images can also be used to summarize; plenty of information can be perfectly summed up in a single graph. Lastly, another reason to use images is comparison. Graphs and diagrams are great tools to indicate the differences between two agents.

Do not fill up your images with too much information because it would complicate the readers’ understanding. Images combine or support the written words, but should not be used to replace them. A good combination of words and images can ease the paper’s general understanding.

Thinking Visually: How to Choose?

It is important to know the possibilities each tool offers. Graphs, for example, are good to express the mathematical relationship or statistical correlation between data. Line graphs are useful to present an evolution, circulant graphs are better to indicate proportional parts and column graphs are commonly used to compare different elements.

Researchers and academics are supposed to have a good command of graphs usage . However, the capacity of selecting which data is most likely to be shown this way makes the difference. Indeed, achieving a good command of these tools is quite difficult, but is possible with experience.

Last but not least, it is always helpful to consider the final goal of an academic paper: communication . Thus, if the graph clearly points to one of the research’s main statements, do not doubt in using it.

Rate this article Cancel Reply

Your email address will not be published.

Enago Academy's Most Popular Articles

• Reporting Research

Explanatory & Response Variable in Statistics — A quick guide for early career researchers!

Often researchers have a difficult time choosing the parameters and variables (like explanatory and response…

• Manuscript Preparation
• Publishing Research

How to Use Creative Data Visualization Techniques for Easy Comprehension of Qualitative Research

“A picture is worth a thousand words!”—an adage used so often stands true even whilst…

• Figures & Tables

Effective Use of Statistics in Research – Methods and Tools for Data Analysis

Remember that impending feeling you get when you are asked to analyze your data! Now…

• Old Webinars
• Webinar Mobile App

SCI中稿技巧: 提升研究数据的说服力

如何寻找原创研究课题 快速定位目标文献的有效搜索策略 如何根据期刊指南准备手稿的对应部分 论文手稿语言润色实用技巧分享，快速提高论文质量

Distill: A Journal With Interactive Images for Machine Learning Research

Research is a wide and extensive field of study. This field has welcomed a plethora…

Explanatory & Response Variable in Statistics — A quick guide for early career…

How to Create and Use Gantt Charts

Sign-up to read more

Subscribe for free to get unrestricted access to all our resources on research writing and academic publishing including:

• 2000+ blog articles
• 50+ Webinars
• 10+ Expert podcasts
• 50+ Infographics
• 10+ Checklists
• Research Guides

We hate spam too. We promise to protect your privacy and never spam you.

I am looking for Editing/ Proofreading services for my manuscript Tentative date of next journal submission:

What should universities' stance be on AI tools in research and academic writing?

Help | Advanced Search

Computer Science > Computer Vision and Pattern Recognition

Title: s2rc-gcn: a spatial-spectral reliable contrastive graph convolutional network for complex land cover classification using hyperspectral images.

Abstract: Spatial correlations between different ground objects are an important feature of mining land cover research. Graph Convolutional Networks (GCNs) can effectively capture such spatial feature representations and have demonstrated promising results in performing hyperspectral imagery (HSI) classification tasks of complex land. However, the existing GCN-based HSI classification methods are prone to interference from redundant information when extracting complex features. To classify complex scenes more effectively, this study proposes a novel spatial-spectral reliable contrastive graph convolutional classification framework named S2RC-GCN. Specifically, we fused the spectral and spatial features extracted by the 1D- and 2D-encoder, and the 2D-encoder includes an attention model to automatically extract important information. We then leveraged the fused high-level features to construct graphs and fed the resulting graphs into the GCNs to determine more effective graph representations. Furthermore, a novel reliable contrastive graph convolution was proposed for reliable contrastive learning to learn and fuse robust features. Finally, to test the performance of the model on complex object classification, we used imagery taken by Gaofen-5 in the Jiang Xia area to construct complex land cover datasets. The test results show that compared with other models, our model achieved the best results and effectively improved the classification performance of complex remote sensing imagery.

Submission history

Access paper:.

• HTML (experimental)
• Other Formats

References & Citations

• Google Scholar
• Semantic Scholar

BibTeX formatted citation

Bibliographic and Citation Tools

Code, data and media associated with this article, recommenders and search tools.

• Institution

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs .

An official website of the United States government

The .gov means it’s official. Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

The site is secure. The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

• Publications
• Account settings

Preview improvements coming to the PMC website in October 2024. Learn More or Try it out now .

• Advanced Search
• Journal List
• Springer Nature - PMC COVID-19 Collection
• PMC10068207

Knowledge Graphs: Opportunities and Challenges

Ciyuan peng.

1 Institute of Innovation, Science and Sustainability, Federation University Australia, Ballarat, 3353 VIC Australia

2 School of Computing Technologies, RMIT University, Melbourne, 3000 VIC Australia

Mehdi Naseriparsa

3 Global Professional School, Federation University Australia, Ballarat, 3353 VIC Australia

Francesco Osborne

4 Knowledge Media Institute, The Open University, Milton Keynes, MK7 6AA UK

With the explosive growth of artificial intelligence (AI) and big data, it has become vitally important to organize and represent the enormous volume of knowledge appropriately. As graph data, knowledge graphs accumulate and convey knowledge of the real world. It has been well-recognized that knowledge graphs effectively represent complex information; hence, they rapidly gain the attention of academia and industry in recent years. Thus to develop a deeper understanding of knowledge graphs, this paper presents a systematic overview of this field. Specifically, we focus on the opportunities and challenges of knowledge graphs. We first review the opportunities of knowledge graphs in terms of two aspects: (1) AI systems built upon knowledge graphs; (2) potential application fields of knowledge graphs. Then, we thoroughly discuss severe technical challenges in this field, such as knowledge graph embeddings, knowledge acquisition, knowledge graph completion, knowledge fusion, and knowledge reasoning. We expect that this survey will shed new light on future research and the development of knowledge graphs.

Introduction

Knowledge plays a vital role in human existence and development. Learning and representing human knowledge are crucial tasks in artificial intelligence (AI) research. While humans are able to understand and analyze their surroundings, AI systems require additional knowledge to obtain the same abilities and solve complex tasks in realistic scenarios (Ji et al. 2021 ). To support these systems, we have seen the emergence of many approaches for representing human knowledge according to different conceptual models. In the last decade, knowledge graphs have become a standard solution in this space, as well as a research trend in academia and industry (Kong et al. 2022 ).

Knowledge graphs are defined as graphs of data that accumulate and convey knowledge of the real world. The nodes in knowledge graphs represent the entities of interest, and the edges represent the relations between the entities (Hogan et al. 2021 ; Cheng et al. 2022a ). These representations utilize formal semantics, which allows computers to process them efficiently and unambiguously. For example, the entity “Bill Gates" can be linked to the entity “Microsoft" because Bill Gates is the founder of Microsoft; thus, they have relationships in the real world.

Due to the great significance of knowledge graphs in processing heterogeneous information within a machine-readable context, a considerable amount of research has been conducted continuously on these solutions in recent years (Dai et al. 2020a ). The proposed knowledge graphs are widely employed in various AI systems recently (Ko et al. 2021 ; Mohamed et al. 2021 ), such as recommender systems, question answering, and information retrieval. They are also widely applied in many fields (e.g., education and medical care) to benefit human life and society (Sun et al. 2020 ; Bounhas et al. 2020 ).

Therefore, knowledge graphs have seized great opportunities by improving the quality of AI systems and being applied to various areas. However, the research on knowledge graphs still faces significant technical challenges. For example, there are major limitations in the current technologies for acquiring knowledge from multiple sources and integrating them into a typical knowledge graph. Thus, knowledge graphs provide great opportunities in modern society. However, there are technical challenges in their development. Consequently, it is necessary to analyze knowledge graphs with respect to their opportunities and challenges to develop a better understanding of knowledge graphs.

To deeply understand the development of knowledge graphs, this survey extensively analyzes knowledge graphs in terms of their opportunities and challenges. Firstly, we discuss the opportunities of knowledge graphs in terms of two aspects: AI systems whose performance is significantly improved by knowledge graphs and application fields that benefit from knowledge graphs. Then, we analyze the challenges of knowledge graphs by considering the limitations of knowledge graph technologies. The main contributions of this paper are as follows:

• Survey on knowledge graphs : We conduct a comprehensive survey of existing knowledge graph studies. In particular, this work thoroughly analyzes the advancements in knowledge graphs in terms of state-of-the-art technologies and applications.
• Knowledge graph opportunities : We investigate potential opportunities for knowledge graphs in terms of knowledge graph-based AI systems and application fields that utilize knowledge graphs. Firstly, we examine the benefits of knowledge graphs for AI systems, including recommender systems, question-answering systems, and information retrieval. Then, we discuss the far-reaching impacts of knowledge graphs on human society by describing current and potential knowledge graph applications in various fields (e.g., education, scientific research, social media, and medical care).
• Knowledge graph challenges : We provide deep insights into significant technical challenges facing knowledge graphs. In particular, we elaborate on limitations concerning five representative knowledge graph technologies, including knowledge graph embeddings, knowledge acquisition, knowledge graph completion, knowledge fusion, and knowledge reasoning.

The rest of the paper is organized as follows. Section  2 provides an overview of knowledge graphs, including the definitions and the categorization of existing research on knowledge graphs. To examine the opportunities of knowledge graphs, Section  3 and Section  4 introduce relevant AI systems and application fields, respectively. Section  5 details the challenges of knowledge graphs based on the technologies. Finally, we conclude this paper in Section  6 .

In this section, the definition of knowledge graphs is provided first. Then, we categorize significant state-of-the-art research in this area.

What Are Knowledge Graphs?

A knowledge base is a typical data set that represents real-world facts and semantic relations in the form of triplets. When the triplets are represented as a graph with edges as relations and nodes as entities, it is considered a knowledge graph. Generally, the knowledge graph and knowledge base are regarded as the same concept and are used interchangeably. In addition, the schema for a knowledge graph can be defined as an ontology, which shows the properties of a specific domain and how they are related. Therefore, one essential stage of knowledge graph construction is ontology construction.

In 2012, Google first put forward Knowledge Graph by introducing their knowledge base called Google Knowledge Graph (Ehrlinger and Wöß 2016 ). Afterward, many knowledge graphs are introduced and adopted such as:

• DBpedia , a knowledge graph that intends to discover semantically meaningful information form Wikipedia and convert it into an effective well-structured ontological knowledge base in DBpedia (Auer et al. 2007 ).
• Freebase , a knowledge graph which is built upon multiple sources that provides a structured and global resource of information (Bollacker et al. 2008 ).
• Facebook’s entity graph , a knowledge graph that converts the unstructured content of the user profiles into meaningful structured data (Ugander et al. 2011 ).
• Wikidata , a cross-lingual document-oriented knowledge graph which supports many sites and services such as Wikipedia (Vrandečić and Krötzsch 2014 ).
• Yago , a quality knowledge base that contains a huge number of entities and their corresponding relationships. These entities are extracted from multiple sources such as Wikipedia and WordNet (Rebele et al. 2016 ).
• WordNet , a lexical knowledge base measuring the semantic similarity between words. The knowledge base contains a number of hierarchical concept graphs to analyse the semantic similarity (Pedersen et al. 2004 ).

A knowledge graph is a directed graph composed of nodes and edges, where one node indicates an entity (a real object or abstract concept), and the edge between the two nodes conveys the semantic relation between the two entities (Bordes et al. 2011 ). Resource Description Framework (RDF) and Labeled Property Graphs (LPGs) are two typical ways to represent and manage knowledge graphs (Färber et al. 2018 ; Baken 2020 ). The fundamental unit of a knowledge graph is the triple (subject, predicate, object) (or (head, relation, tail) ), i.e., (Bill Gates, founderOf, Microsoft) . Since the relation is not necessarily symmetric, the direction of a link matters. Therefore, a knowledge graph can also be seen as a directed graph in which the head entities point to the tail entities via the relation’s edge.

Fig.  1 depicts an example of a simple knowledge graph. As shown in Fig.  1 , nodes e 1 and e 2 darkened in color are connected by relation r 1 , which goes from e 1 to e 2 . Therefore, e 1 , e 2 , and r 1 can form the triplet ( e 1 , r 1 , e 2 ) , in which e 1 and e 2 are the head and tail entities, respectively.

An example of a knowledge graph. In this knowledge graph, ( e 1 , r 1 , e 2 ) is a triplet that indicates e 1 and e 2 are connected by relation r 1

Current Research on Knowledge Graphs

In recent years, knowledge graphs have gained extensive research interest. Plenty of studies have focused on exploring knowledge graphs. This paper conducts a comprehensive survey on knowledge graphs and lists seven important categories of current research on this topic. Fig.  2 illustrates a schema of the most popular research lines regarding knowledge graphs. Among them, AI systems are services that utilize knowledge graphs for their foundation, and application fields are domains where knowledge graphs reach. These two research lines are listed for discussing the opportunities of knowledge graphs. Another five research lines are five main knowledge graph technologies corresponding to five tasks. In this paper, we introduce these five technologies and emphasize their limitations to give useful insights into the major challenges of the knowledge graphs.

Research on knowledge graphs

Knowledge Graph Embedding

Knowledge graph embedding is one of the central research issues. This task aims to map entities and relations of a knowledge graph to a low-dimensional vector space so that it captures the semantics and the structure of the knowledge graph efficiently (Dai et al. 2020b ). Then, the obtained feature vectors can be effectively learned by machine learning models. Three main triplet fact-based embedding methods are as follows: (a) tensor factorization-based, (b) translation-based, and (c) neural network-based methods (Dai et al. 2020b ).

Knowledge Acquisition

Knowledge acquisition, which focuses on modeling and constructing knowledge graphs, is another crucial research direction of knowledge graph study. Typically, the knowledge is imported from structured sources by employing mapping languages, such as R2RML (Rodriguez-Muro and Rezk 2015 ). Furthermore, the knowledge could be extracted from unstructured documents (e.g., news, research papers, and patents) by adopting relation, entity, or attribute extraction methods (Liu et al. 2020 ; Yu et al. 2020 ; Yao et al. 2019 ).

Knowledge Graph Completion

Although there are many methods for constructing knowledge graphs, it is still unfeasible to create comprehensive representations of all the knowledge in a field. Most knowledge graphs still lack a good number of entities and relationships. Thereby, significant efforts have been made for completing knowledge graphs. Knowledge graph completion aims to improve the quality of knowledge graphs by predicting additional relationships and entities. The first task typically adopts link prediction techniques to generate triplets and then assigns the triplets plausibility scores (Ji et al. 2021 ). The second task employs entity prediction methods for obtaining and integrating further information from external sources.

Knowledge Fusion

Knowledge fusion is also an important research direction that focuses on capturing knowledge from different sources and integrating it into a knowledge graph (Nguyen et al. 2020 ). The knowledge fusion approaches are useful for both generating and completing knowledge graphs. Recently, entity alignment has been the primary method for implementing knowledge fusion tasks.

Knowledge Reasoning

Tremendous research efforts have focused on reasoning to enrich the knowledge graphs, which aims to infer new facts based on existing data (Minervini et al. 2020 ). In particular, new relations between two unconnected entities are inferred, forming new triplets. Also, by reasoning out the false facts, knowledge reasoning has the ability to identify erroneous knowledge. The main methods for knowledge reasoning include logic rule-based, distributed representation-based, and neural network-based methods (Chen et al. 2020b ).

Nowadays, knowledge graphs are widely utilized by AI systems (Liang et al. 2022 ), such as recommenders, question-answering systems, and information retrieval tools. Typically, the richness of information within knowledge graphs enhances the performance of these solutions. Therefore, many studies have focused on taking advantage of knowledge graphs to improve AI systems’ performance.

Application Fields

Knowledge graphs have numerous applications in various fields, including education, scientific research, social media, and medical care (Li et al. 2020b ). A variety of intelligent applications are required to improve the standard of human life.

Differing from other works, this paper focuses on surveying the opportunities and challenges of knowledge graphs. In particular, knowledge graphs meet great opportunities by improving the quality of AI services and being applied in various fields. On the contrary, this paper regards the limitations of knowledge graph technologies as the challenges. Therefore, we will discuss the technical limitations regarding knowledge graph embeddings, knowledge acquisition, knowledge graph completion, knowledge fusion, and knowledge reasoning.

Knowledge Graphs for AI Systems

This section explains the opportunities by analyzing the advantages that knowledge graphs bring for improving the functionalities of AI Systems. Specifically, there are a couple of systems, including recommender systems, question-answering systems, and information retrieval tools (Guo et al. 2020 ; Zou 2020 ), which utilize knowledge graphs for their input data and benefit the most from knowledge graphs. In addition to these systems, other AI systems, such as image recognition systems (Chen et al. 2020a ), have started to consider the characteristic of knowledge graphs. However, the application of knowledge graphs in these systems is not widespread. Moreover, these systems do not directly optimize performance by utilizing knowledge graphs as input data. Therefore, the advantages that knowledge graphs bring for recommender systems, question-answering systems, and information retrieval tools are discussed in detail to analyze the opportunities of knowledge graphs. Typically, these solutions greatly benefit from adopting knowledge graphs that offer high-quality representations of the domain knowledge. Table  1 presents a summary of the AI systems that we will discuss below.

AI systems using knowledge graphs

Recommender Systems

With the continuous development of big data, we observe the exponential growth of information. In the age of information explosion, it becomes challenging for people to receive valid and reliable information (Shokeen and Rana 2020 ; Monti et al. 2021 ; Gómez et al. 2022 ). Specifically, online users may feel confused when they want to select some items they are interested in among thousands of choices. To tackle this issue, we saw the emergence of several recommender systems to provide users with more accurate information. Typically, recommender systems learn the preference of target users for a set of items (Wan et al. 2020 ; Zheng and Wang 2022 ) and produce a set of suggested items with similar characteristics. Recommender systems are fruitful solutions to the information explosion problem and are employed in various fields for enhancing user experience (Quijano-Sánchez et al. 2020 ).

Traditional Recommender Systems

There are two traditional methods for developing recommender systems, including content-based and collaborative filtering-based (CF-based) methods. Sun et al. ( 2019 ) and Guo et al. ( 2020 ) have compared and summarised these two approaches.

Content-Based Recommender Systems

The content-based recommender systems first analyze the content features of items (e.g., descriptions, documents). These items are previously scored by the target users (Guo et al. 2020 ; Xia et al. 2014b ). Then, the recommender systems learn the user interests by employing machine learning models. Thus, these systems are able to effectively recommend trending items to the target users according to their preferences. Some recommender systems utilize the content of the original query result to discover highly-related items for the users that may interest them (Naseriparsa et al. 2019a ). These systems employ machine learning techniques or statistical measures such as correlation to compute the highly-similar items to those that are visited by the users (Naseriparsa et al. 2019b ). Another group of content-based recommender systems employs lexical references such as dictionaries to utilize semantic relationships of the user query results to recommend highly semantically-related items to the users that may directly satisfy their information needs (Naseriparsa et al. 2018 ; Sun et al. 2017 ).

CF-Based Recommender Systems

CF-based recommender systems suggest items to the users based on the information of user-item interaction (Chen et al. 2020c ). CF-based recommender systems infer the user preference by clustering similar users instead of extracting the features of the items (Wang et al. 2019a ). However, we face data sparsity and cold start problems in traditional CF-based systems. In general, users can only rate a few items among a large number of items, which leads to preventing many items from receiving appropriate feedback. Therefore, the recommender systems do not effectively learn user preferences accurately because of data sparsity (Bai et al. 2019 ; Xia et al. 2014a ). On the other hand, the cold start problem makes it even more difficult to make recommendations when the items or users are new because there is no historical data or ground truth. Moreover, because abundant user information is required for achieving effective recommendations, CF-based recommender systems face privacy issues. How to achieve personalized recommendations while protecting the privacy of users is still an unsolved problem.

Knowledge Graph-Based Recommender Systems

To address inherent problems of traditional approaches, the community has produced several hybrid recommender systems, which consider both item features and the distribution of user scores. Most of these solutions adopt knowledge graphs for representing and interlinking items (Palumbo et al. 2020 ). Specifically, Knowledge graph-based recommender systems integrate knowledge graphs as auxiliary information and leverage users and items networks to learn the relationships of items-users, items-items, and users-users (Palumbo et al. 2018 ).

Fig  3 presents an example of knowledge graph-based movie recommendation. Here we can see that the movies “Once Upon A Time in Hollywood" and “Interstellar" are recommended to three users according to a knowledge graph that contains the nodes of users, films, directors, actors, and genres. The knowledge graph is thus used to infer latent relations between the user and the recommended movies.

An example of knowledge graph-based recommender system

Recently, a great deal of research has been conducted to utilize knowledge graphs for recommendation tasks. For instance, Wang et al. ( 2019b ) introduced KPRN. KPRN is a recommender system that generates entity-relation paths according to the user-item interaction and constructs a knowledge graph that consists of the users, items, and their interaction. It then infers the user preference based on the entity-relation path. The user-item interaction, which is extracted from knowledge graphs, improves the quality of the recommendations and allows the presentation of the recommended results in a more explainable manner. Wang et al. ( 2019c ) also applied multi-task knowledge graph representation (MKR) for recommendation tasks. MKR models knowledge graphs based on the user-item interaction. It is worth noting that MKR focuses on the structural information of knowledge graphs for learning the latent user-item interaction. Sun et al. ( 2020 ) proposed a Multi-modal Knowledge Graph Attention Network (MKGAT) for achieving precise recommendations. MKGAT constructs knowledge graphs based on two aspects: (1) it enriches entity information by extracting the information of the neighbor entities; (2) it scores the triplets to construct the reasoning relations. Finally, they applied knowledge graphs that are enriched with structured data to recommender systems.

Wang et al. ( 2018b ) presented the RippleNet model, which incorporates knowledge graphs into recommendation tasks by preference propagation. RippleNet firstly regards users’ historical records as the basis of a knowledge graph. Then, it predicts the user preference list among candidate items based on the knowledge graph links. Based on both RippleNet and MKR models, Wang et al. ( 2021 ) applied the Ripp-MKR model. Ripp-MKR combines the advantages of preference propagation and user-item interaction to dig the potential information of knowledge graphs. Shu and Huang ( 2021 ) proposed RKG, which achieves recommendation by referring to the user preference-based knowledge graph. RKG first obtains users’ preference lists; then, it analyzes the relations between the user’s preferred items and the items which are to be recommended. Therefore, the model effectively learns the scores of the candidate items according to the relationships between candidate items and the user’s preferred items.

Many studies have utilized ontological knowledge base information to improve retrieving results from various data sources (Farfán et al. 2009 ). Wu et al. ( 2013 ) adopted the ontological knowledge base to extract highly semantically similar sub-graphs in graph databases. Their method effectively recommends semantically relevant sub-graphs according to ontological information. Farfán et al. ( 2009 ) proposed the XOntoRank, which adopts the ontological knowledge base to facilitate the data exploration and recommendation on XML medical records.

Compared with the traditional recommender systems, knowledge graph-based recommender systems have the following advantages:

• Better Representation of Data: Generally, the traditional recommender systems suffer from data sparsity issues because users usually have experience with only a small number of items. However, the rich representation of entities and their connections in knowledge graphs alleviate this issue.
• Alleviating Cold Start Issues: It becomes challenging for traditional recommender systems to make recommendations when there are new users or items in the data set. In knowledge graph-based recommender systems, information about new items and users can be obtained through the relations between entities within knowledge graphs. For example, when a new Science-Fiction movie such as “Tenet” is added to the data set of a movie recommender system that employs knowledge graphs, the information about “Tenet" can be gained by its relationship with the genre Science-Fiction (gaining triplet (Tenet, has genre of, Sci-Fi) ).
• The Explainability of Recommendation: Users and the recommended items are connected along with the links in knowledge graphs. Thereby, the reasoning process can be easily illustrated by the propagation of knowledge graphs.

Question–Answering Systems

Question answering is one of the most central AI services, which aims to search for the answers to natural language questions by analyzing the semantic meanings (Dimitrakis et al. 2020 ; Das et al. 2022 ). The traditional question-answering systems match the textual questions with the answers in the unstructured text database. In the search process, the semantic relationship between the question and answer is analyzed; then, the system matches the questions and answers with the maximum semantic similarity. Finally, the system outputs the answer. However, the answers are obtained by filtrating massive unstructured data, which deteriorates the efficiency of the traditional question-answering systems due to analyzing an enormous search space. To solve this issue, a lot of research focuses on employing structured data for question answering, particularly knowledge graph-based question-answering systems (Singh et al. 2020 ; Qiu et al. 2020 ).

The sophisticated representation of information in knowledge graphs is a natural fit for question-answering systems. Knowledge graph-based question-answering systems typically analyze the user question and retrieve the portion of knowledge graphs for answering. The answering task is facilitated either by using similarity measures or by producing structured queries in standard formats (e.g., SPARQL). Fig  4 presents an example of the knowledge graph-based question-answering system. The system answer “Shakespeare" is a node that is linked to the node “Romeo". The node “Romeo" is extracted from the question.

The illustration of knowledge graph based question-anwsering systems

There are two main types of questions in this space: simple and multi-hop questions, respectively. Simple questions are answered only by referring to a single triplet, while multi-hop questions require combining multiple entities and relations. Focusing on simple questions, Huang et al. ( 2019 ) proposed a knowledge graph embedding-based question-answering system (KEQA). They translated the question and its corresponding answer into a single triplet. For instance, the question “ Which film acted by Leonardo" and one of its answers “Inception" can be expressed as the following triplet: ( Leonard ,  act ,  Inception ). Then, the head entity, relation, and tail entity of the triplet are represented by a vector matrix in the embedding space for learning the question-answer information. Considering the semantic meanings of the questions, Shin et al. ( 2019 ) presented a predicate constraint-based question-answering system (PCQA). They took advantage of the predicate constraints of knowledge graphs, which is a triplet contains a subject, predicate, and an object to capture the connection between the questions and answers. Using the triplet for question-answering integration, the processing of the question-answering service can be simplified; therefore, the result improves.

Bauer et al. ( 2018 ) focused on multi-hop questions and proposed a Multi-Hop Pointer-Generator Model (MHPGM). They selected the relation edges that are related to the questions in a knowledge graph and injected attention to achieve multi-hop question answering. Because of the advantages of knowledge graphs’ structure, multi-hop question answering can extract coherent answers effectively. Saxena et al. ( 2020 ) proposed EmbedKGQA to achieve multi-hop question answering over sparse knowledge graphs (such as knowledge graphs with missing edges). The main idea of EmbedKGQ is to utilize knowledge graph embeddings to reduce knowledge graph sparsity. It first creates embeddings of all entities and then selects the embedding of a given question. Lastly, it predicts the answer by combining these embeddings.

Compared to the traditional question answering, the advantages of knowledge graph-based question-answering systems can be summarized as follows:

• Increased Efficiency: Instead of searching for answers from massive textual data, which may contain a large volume of useless data items, knowledge graph-based question-answering systems focus only on entities with relevant properties and semantics. Therefore, they reduce the search space significantly and extract the answers effectively and efficiently.
• Multi-hop Question Answering: The answers can be more complex and sophisticated than the ones produced with traditional methods since facts and concepts from knowledge graphs can be combined via multi-hop question answering.

Information Retrieval

Information retrieval enables retrieval systems to match end-user queries with relevant documents, such as web pages (Liu et al. 2019 ). Traditional information retrieval systems index the documents according to the user queries and return the matched documents to the users (Hersh 2021 ). Nevertheless, index processing is complex and requires plenty of time because of the massiveness and diversity of documents. As a result, traditional information retrieval faces the challenge of inaccurate search results and potentially low efficiency. Also, since search engines have limitations with respect to text interpretation ability, keyword-based text search usually outputs limited results. Thus, to address these problems, many modern search engines take advantage of knowledge graphs (Bounhas et al. 2020 ; Zheng et al. 2020 ). Knowledge graph-based information retrieval introduces a new research direction that takes advantage of knowledge graphs for improving the performance of search engines and the explainability of the results.

Typically, these systems rely on the advanced representation of the documents based on entities and relationships from knowledge graphs. These formal and machine-readable representations are then matched to the user query for retrieving the more pertinent documents. For instance, Wise et al. ( 2020 ) proposed a COVID-19 Knowledge Graph (CKG) to extract the relationships between the scientific articles about COVID-19. In particular, they combined the topological information of documents with the semantic meaning to construct document knowledge graphs. Wang et al. ( 2018a ) proposed a knowledge graph-based information retrieval technology that extracts entities by mining entity information on web pages via an open-source relation extraction method. Then, the entities with relationships are linked to construct a knowledge graph.

Knowledge graphs can also support methods for query expansion, which is able to enrich the user query by adding relevant concepts (e.g., synonymous). For example, Dalton et al. ( 2014 ) presented an entity query feature expansion (EQFE) to enrich the queries based on the query knowledge graph, including structured attributes and text. Liu et al. ( 2018 ) proposed the Entity-Duet Neural Ranking Model (EDRM). EDRM integrates the semantics extracted from knowledge graphs with the distributed representations of entities in queries and documents. Then, it ranks the search results using interaction-based neural ranking networks.

Compared to traditional information retrieval, the knowledge graph-based information retrieval has the following advantages:

• Semantic Representation of Items: Items are represented according to a formal and interlinked model that supports semantic similarity, reasoning, and query expansion. This typically allows the system to retrieve more relevant items and makes the system more interpretable.
• High Search Efficiency: Knowledge graph-based information retrieval can use the advanced representation of the items to reduce the search space significantly (e.g., discarding documents that use the same terms with different meanings), resulting in improved efficiency.
• Accurate Retrieval Results: In knowledge graph-based information retrieval, the correlation between query and documents is analyzed based on the relations between entities in the knowledge graph. This is more accurate than finding the similarities between queries and documents.

Applications and Potentials

In this section, we discuss the applications and potentials of knowledge graphs in four domains: education, scientific research, social networks, and health/medical care. Although some researchers try to take advantage of knowledge graphs to develop beneficial applications in other domains such as finance (Cheng et al. 2022a ), the knowledge graph-based intelligent service in these areas is relatively obscure and still needs to be explored. Therefore, this section mainly focuses on education, scientific research, social networks, and medical care to summarize the opportunities of knowledge graphs. Table   2 presents several recent applications of knowledge graphs that make contributions to these fields.

Fields of applications of knowledge graphs

Education is of great importance to the development of human society. Many studies have focused on deploying intelligent applications to improve the quality of education (Bai et al. 2021 ; Wang et al. 2020c ). Specifically, in the age of big data, data processing becomes a challenging task because of the complex and unstructured educational data. Thereby, intelligent educational systems tend to apply structured data, such as knowledge graphs. Several knowledge graph-based applications support the educational process, focusing in particular on data processing and knowledge dissemination (Yao et al. 2020 ).

In education, the quality of offline school teaching is of vital importance. Therefore, several knowledge graph-based applications focus on supporting teaching and learning. For example, considering the importance of course allocation tasks in university, Aliyu et al. ( 2020 ) proposed a knowledge graph-based course management approach to achieve automatic course allocation. They constructed a course knowledge graph in which the entities are courses, lecturers, course books, and authors in order to suggest relevant courses to students. Chen et al. ( 2018 ) presented KnowEdu, a system for educational knowledge graph construction, which automatically builds knowledge graphs for learning and teaching in schools. First, KnowEdu extracts the instructional concepts of the subjects and courses as the entity features. Then, it identifies the educational relations based on the students’ assessments and activities to make the teaching effect more remarkable.

The abovementioned knowledge graph-based intelligent applications are dedicated to improving the quality of offline school teaching. However, online learning has become a hot trend recently. Moreover, online study is an indispensable way of learning for students during the COVID-19 pandemic(Saraji et al. 2022 ). Struggling with confusing online content (e.g., learning content of low quality on social media), students face major challenges in acquiring significant knowledge efficiently. Therefore, researchers have focused on improving online learning environments by constructing education-efficient knowledge graphs (d’Aquin 2016 ; Pereira et al. 2017 ). For example, to facilitate online learning and establish connections between formal learning and social media, Zablith ( 2022 ) proposed to construct a knowledge graph by integrating social media and formal educational content, respectively. Then, the produced knowledge graph can filter social media content, which is fruitful for formal learning and help students with efficient online learning to some extent.

Offline school teaching and online learning are two essential parts of education, and it is necessary to improve the quality of both to promote the development of education. Significantly, knowledge graph-based intelligent applications can deal with complicated educational data and make both offline and online education more convenient and efficient.

Scientific Research

A variety of knowledge graphs focus on supporting the scientific process and assisting researchers in exploring research knowledge and identifying relevant materials (Xia et al. 2016 ). They typically describe documents (e.g., research articles, patents), actors (e.g., authors, organizations), entities (e.g., topics, tasks, technologies), and other contextual information (e.g., projects, funding) in an interlinked manner. For instance, Microsoft Academic Graph (MAG) (Wang et al. 2020a ) is a heterogeneous knowledge graph. MAG contains the metadata of more than 248M scientific publications, including citations, authors, institutions, journals, conferences, and fields of study. The AMiner Graph (Zhang et al. 2018 ) is the corpus of more than 200M publications generated and used by the AMiner system 1 . The Open Academic Graph (OAG) 2 is a massive knowledge graph that integrates Microsoft Academic Graph and AMiner Graph. AceKG (Wang et al. 2018c ) is a large-scale knowledge graph that provides 3 billion triples of academic facts about papers, authors, fields of study, venues, and institutes, as well as the relations among them. The Artificial Intelligence Knowledge Graph (AI-KG) (Dessì et al. 2020 ) 3 describes 800K entities (e.g., tasks, methods, materials, metrics) extracted from the 330K most cited articles in the field of AI. The Academia/Industry Dynamics Knowledge Graph (AIDA KG) (Angioni et al. 2021 ) 4 describes 21M publications and 8M patents according to the research topics drawn from the Computer Science Ontology (Salatino et al. 2020 ) and 66 industrial sectors (e.g., automotive, financial, energy, electronics).

In addition to constructing academic knowledge graphs, many researchers also take advantage of knowledge graphs to develop various applications beneficial to scientific research. Chi et al. ( 2018 ) proposed a scientific publication management model to help non-researchers learn methods for sustainability from research thinking. They built a knowledge graph-based academic network to manage scientific entities. The scientific entities, including researchers, papers, journals, and organizations, are connected regarding their properties. For the convenience of researchers, many scientific knowledge graph-based recommender systems, including citation recommendation, collaboration recommendation, and reviewer recommendation, are put forward (Shao et al. 2021 ). For instance, Yong et al. ( 2021 ) designed a knowledge graph-based reviewer assignment system to achieve precise matching of reviewers and papers. Particularly, they matched knowledge graphs and recommendation rules to establish a rule engine for the recommendation process.

Social Networks

With the rapid growth of social media such as Facebook and Twitter, online social networks have penetrated human life and bring plenty of benefits such as social relationship establishment and convenient information acquisition (Li et al. 2020a ; Hashemi and Hall 2020 ). Various social knowledge graphs are modeled and applied to analyze the critical information from the social network. These knowledge graphs are usually constituted based on the people’s activities and their posts on social media, which are applied to numerous applications for different functions (Xu et al. 2020 ).

Remarkably, social media provides high chances for people to make friends and gain personalized information. Furthermore, social media raises fundamental problems, such as how to recommend accurate content that interests us and how to connect with persons interested in a common topic. To address these issues, various studies have been proposed to match users with their favorite content (or friends) for recommendation (Ying et al. 2018 ). With the increase in users’ demand, a number of researchers utilize knowledge graph-based approaches for more precise recommendations (Gao et al. 2020 ). A representative example is GraphRec (a graph neural network framework for social recommendations) proposed by Fan et al. ( 2019 ). They considered two kinds of social knowledge graphs: user-user and user-item graphs. Then, they extracted information from the two knowledge graphs for the learning task. As a result, their model can provide accurate social recommendations because it aggregates the social relationships of users and the interactions between users and items.

In addition, people’s activities on social media reveal social relationships. For example, we can learn about the relationships around a person through his photos or comments on Twitter. Significantly, social relationship extraction assists companies in tracking users and enhancing the user experience. Therefore, many works are devoted to social relationship extraction. Wang et al. ( 2018d ) propose a graph reasoning model to recognize the social relationships of people in a picture that is posted on social media. Their model enforces a particular function based on the social knowledge graph and deep neural networks. In their method, they initialized the relation edges and entity nodes with the features that are extracted from the semantic objects in an image. Then, they employed GGNN to propagate the knowledge graph. Therefore, they explored the relations of the people in the picture.

One of the biggest problems in this space is fake news (Zhang et al. 2019a ). Online social media has become the principal platform for people to consume news. Therefore, a considerable amount of research has been done for fake news detection (Choi et al. 2020 ; Meel and Vishwakarma 2020 ). Most recently, Mayank et al. ( 2021 ) exploited a knowledge graph-based model called DEAP-FAKED to detect fake news on social media. Specifically, DEAP-FAKED learns news content and identifies existing entities in the news as the nodes of the knowledge graph. Afterward, a GNN-based technique is applied to encode the entities and detect anomalies that may be linked with fake news.

Health/Medical Care

With medical information explosively growing, medical knowledge analysis plays an instrumental role in different healthcare systems. Therefore, research focuses on integrating medical information into knowledge graphs to empower intelligent systems to understand and process medical knowledge quickly and correctly (Li et al. 2020b ). Recently, a variety of biomedical knowledge graphs have become available. Therefore, many medical care applications exploit knowledge graphs. For instance, Zhang et al. ( 2020a ) presented a Health Knowledge Graph Builder (HKGB) to build medical knowledge graphs with clinicians’ expertise.

Specifically, we discuss the three most common intelligent medical care applications, including medical recommendation, health misinformation detection, and drug discovery. Firstly, with the rapid development of the medical industry, medical choices have become more abundant. Nevertheless, in the variety of medical choices, people often feel confused and unable to make the right decision to get the most suitable and personalized medical treatment. Therefore, medical recommender systems, especially biomedical knowledge graph-based recommender systems (such as doctor recommender systems and medicine recommender systems), have been put forward to deal with this issue (Katzman et al. 2018 ). Taking medicine recommendation as an example, Gong et al. ( 2021 ) provided a medical knowledge graph embedding method by constructing a heterogeneous graph whose nodes are medicines, diseases, and patients to recommend accurate and safe medicine prescriptions for complicated patients.

Secondly, although many healthcare platforms aim to provide accurate medical information, health misinformation is an inevitable problem. Health misinformation is defined as incorrect information that contradicts authentic medical knowledge or biased information that covers only a part of the facts (Wang et al. 2020d ). Unfortunately, a great deal of health-related information on various healthcare platforms (e.g., medical information on social media) is health misinformation. What’s worse, the wrong information leads to consequential medical malpractice; therefore, it is urgent to detect health misinformation. Utilizing authoritative medical knowledge graphs to detect and filter misinformation can help people make correct treatment decisions and suppress the spread of misinformation (Cui et al. 2020 ). Representatively, Cui et al. ( 2020 ) presented a model called DETERREN to detect health misinformation. DETERREN leverages a knowledge-guided attention network that incorporates an article-entity graph with a medical knowledge graph.

Lastly, drug discovery, such as drug repurposing and drug-drug interaction prediction, has been a research trend for intelligent healthcare in recent years. Benefiting from the rich entity information (e.g., the ingredients of a drug) and relationship information (e.g., the interaction of drugs) in medical knowledge graphs, drug discovery based on knowledge graphs is one of the most reliable approaches (MacLean 2021 ). Lin et al. ( 2020 ) presented an end-to-end framework called KGNN (Knowledge Graph Neural Network) for drug-drug interaction prediction. The main idea of KGNN is to mine the relations between drugs and their potential neighborhoods in medical knowledge graphs. It first exploits the topological information of each entity; then, it aggregates all the neighborhood information from the local receptive entities to extract both semantic relations and high-order structures. Wang et al. ( 2020e ) developed a knowledge discovery framework called COVID-KG to generate COVID-19-related drug repurposing reports. They first constructed multimedia knowledge graphs by extracting medicine-related entities and their relations from images and texts. Afterward, they utilized the constructed knowledge graphs to generate drug repurposing reports.

Technical Challenges

Although knowledge graphs offer fantastic opportunities for various services and applications, many challenges are yet to be addressed (Noy et al. 2019 ). Specifically, the limitations of existing knowledge graph technologies are the key challenges for promoting the development of knowledge graphs (Hogan et al. 2021 ). Therefore, this section discusses the challenges of knowledge graphs in terms of the limitations of five topical knowledge graph technologies, including knowledge graph embeddings, knowledge acquisition, knowledge graph completion, knowledge fusion, and knowledge reasoning.

Knowledge Graph Embeddings

The aim of knowledge graph embeddings is to effectively represent knowledge graphs in a low-dimensional vector space while still preserving the semantics (Xia et al. 2021 ; Vashishth et al. 2020 ). Firstly, the entities and relations are embedded into a dense dimensional space in a given knowledge graph, and a scoring function is defined to measure the plausibility of each fact (triplet). Then, the plausibility of the facts is maximized to obtain the entity and relation embeddings (Chaudhri et al. 2022 ; Sun et al. 2022 ). The representation of knowledge graphs brings various benefits to downstream tasks. The three main types of triplet fact-based knowledge graph embedding approaches are tensor factorization-based, translation-based, and neural network-based methods (Rossi et al. 2021 ).

Tensor Factorization-Based Methods

The core idea of tensor factorization-based methods is transforming the triplets in the knowledge graph into a 3D tensor (Balažević et al. 2019 ). As Fig  5 presents, the tensor X ∈ R m × m × n , where m and n indicate the number of entity and relation, respectively, contains n slices, and each slice corresponds to one relation type. If the condition X ijk = 1 is met, the triplet ( e i , r k , e j ) , where e and r denote entity and relation, respectively, exists in the knowledge graph. Otherwise, if X ijk = 0 , there is no such a triplet in the knowledge graph. Then, the tensor is represented by the embedding matrices that consist of the vectors of entities and relations.

An illustration of tensor factorization of knowledge graphs

Translation-Based Methods

Translation-based methods exploit the scoring function, which is based on translation invariance. Translation invariance interprets the distance between the vectors of the two words, which is represented by the vector of their semantic relationships (Mikolov et al. 2013 ). Bordes et al. ( 2013 ) firstly utilized the translation invariance-based scoring functions to measure the embedding results. They creatively proposed the TransE model, which translates all the entities and relations of a knowledge graph into a continuous and low vector space. Specifically, the vectors of the head and tail entities in a triplet are connected by the vector of their relation. Consequently, in the vector space, the semantic meaning of every triplet is preserved. Formally, given a triplet ( head ,  relation ,  tail ), the embedding vectors of the head entity, relation, and tail entity are h , r , and t , respectively. In the vector space, the plausibility of the triplet ( h , r , t ) is computed by the translation invariance-based scoring function to ensure it follows the geometric principle: h + r ≈ t .

After TransE, a lot of related extensions, such as TransH (Wang et al. 2014 ) and TransR (Lin et al. 2015 ), are continually proposed to improve the performance of the Translation-based knowledge graph embeddings.

Neural Network-Based Methods

Nowadays, deep learning has become a popular tool that is utilized for knowledge graph embeddings, and a considerable amount of research proposes to employ neural networks to represent the triplets of knowledge graphs (Dai et al. 2020a ). In this section, we discuss three representative works, including SME, ConvKB, and R-GCN, to briefly introduce neural network-based knowledge graph embeddings.

SME (Bordes et al. 2014 ) designs an energy function to conduct semantic matching, which utilizes neural networks to measure the confidence of each triplet ( h ,  r ,  t ) in knowledge graphs. The scoring function of SME is defined as follows:

The scoring function of SME (bilinear) is:

Here W ∈ R d × d denotes the weight matrix, b indicates the bias vector. h , r , and t are the embedding vectors of head entity, relation, and tail entity, respectively.

ConvKB (Nguyen et al. 2017 ) utilizes a convolutional neural network (CNN) to conduct knowledge graph embeddings. ConvKB represents each triplet ( h ,  r ,  t ) as a three-row matrix A , which is input to a convolution layer to obtain feature maps. Afterward, the feature maps are concatenated as a vector, and then a score is calculated to estimate the confidence of the triplet. The scoring function is as follows:

where O signifies the concatenation operator, g ( · ) is the ReLU activation function, A ∗ Ω indicates the convolution operation of matrix A by using the filters in the set Ω , w ∈ R 3 d is a weight vector.

R-GCN (Schlichtkrull et al. 2018 ) is an improvement of graph neural networks (GNNs). R-GCN represents knowledge graphs by providing relation-specific transformation. Its forward propagation is calculated as follows:

where h k ( l + 1 ) is the hidden state of the entity k in l -th layer, N k r denotes a neighbor collection of entity k and relation r ∈ R , n k , r is the normalization process, W i ( l ) and W k ( l ) are the weight matrices.

Limitations of Existing Methods

The existing methods for generating knowledge graph embeddings still suffer several severe limitations. Many established methods only consider surface facts (triplets) of knowledge graphs. However, additional information, such as entity types and relation paths, are ignored, which can further improve the embedding accuracy. The performance of most traditional methods that do not consider the additional information is unsatisfactory. Table  3 lists the embedding methods, which do not consider the additional information. In Table  3 , the performance evaluation is based on the link prediction and triplet classification tasks. The metrics that are for evaluation results are hit rate at 10 (Hits@10) and accuracy. As Table  3 presents, only a few models have impressive results, including the results of QuatE (90%), RMNN (89.9%), and KBGAN (89.2%). Recently, some researchers have started to combine additional information with a knowledge graph to improve the efficiency of embedding models. For example, Guo et al. ( 2015 ) take advantage of additional entity type information, which is the semantic category of each entity, to obtain the correlation between the entities and to tackle the data sparsity issue. Therefore, knowledge graphs are represented more accurately. Not only entity types, some other information, including relation paths (Li et al. 2021 ), time information of dynamic graphs (Messner et al. 2022 ), and textual descriptions of entities (An et al. 2018 ), are getting the researchers’ attention in recent years. However, it is still a daunting challenge to effectively utilize rich additional information to improve the accuracy of knowledge graph embeddings.

Knowledge graph embedding methods

In this table, all the results of link prediction are filter results

General additional information can not adequately represent the semantic meaning of the triplets. For instance, the entity types are not related to the semantic information of triplets. Furthermore, the types of additional information that can be incorporated into the features of the triplets are now severely limited. Therefore, to improve the performance of existing knowledge graph embedding methods, multivariate information (such as the hierarchical descriptions of relations and the combination of entity types and textual descriptions) needs to be incorporated into the features of the triplets.

To the best of our knowledge, complex relation path remains an open research problem (Peng et al. 2021 ). For example, the inherent relations, referring to the indirect relationships between two unconnected entities, are not represented effectively. Although the inherent relations between the entities can be explored based on the chain of relationships in knowledge graphs, the inherent relations are complex and multiple. Therefore, it is not straightforward to represent these relations effectively.

Knowledge acquisition is a critical step for combining data from different sources and generating new knowledge graphs. The knowledge is extracted from both structured and unstructured data. Three main methods of knowledge acquisition are relation extraction, entity extraction, and attribute extraction (Fu et al. 2019 ). Here, attribute extraction can be regarded as a special case of entity extraction. Zhang et al. ( 2019b ) took advantage of knowledge graph embeddings and graph convolution networks to extract long-tail relations. Shi et al. ( 2021 ) proposed entity set expansion to construct large-scale knowledge graphs.

Nevertheless, existing methods for knowledge acquisition still face the challenge of low accuracy, which could result in incomplete or noisy knowledge graphs and hinder the downstream tasks. Therefore, the first critical issue regards the reliability of knowledge acquisition tools and their evaluation. In addition, a domain-specific knowledge graph schema is knowledge-oriented, while a constructed knowledge graph schema is data-oriented for covering all data features (Zhou et al. 2022 ). Therefore, it is inefficient to produce domain-specific knowledge graphs by extracting entities and properties from raw data. Hence, it is an essential issue to efficiently achieve knowledge acquisition tasks by generating domain-specific knowledge graphs.

Besides, most existing knowledge acquisition methods focus on constructing knowledge graphs with one specific language. However, in order to make the information in knowledge graphs richer and more comprehensive, we need cross-lingual entity extraction. It is thus vitally important to give more attention to cross-lingual entity extraction and the generation of multilingual knowledge graphs. For example, Bekoulis et al. ( 2018 ) proposed a joint neural model for cross-lingual (English and Dutch) entity and relation extraction. Nevertheless, multilingual knowledge graph construction is still a daunting task since non-English training data sets are limited, language translation systems are not always accurate, and the cross-lingual entity extraction models have to be retrained for each new language.

Multi-modal knowledge graph construction is regarded as another challenging issue of knowledge acquisition. The existing knowledge graphs are mostly represented by pure symbols, which could result in the poor capability of machines to understand our real world (Zhu et al. 2022b ). Therefore, many researchers focus on multi-modal knowledge graphs with various entities, such as texts and images. The construction of multi-modal knowledge graphs requires the exploration of entities with different modalities, which makes the knowledge acquisition tasks complicated and inefficient.

Knowledge graphs are often incomplete, i.e., missing several relevant triplets and entities (Zhang et al. 2020a ). For instance, in Freebase, one of the most well-known knowledge graphs, more than half of person entities do not have information about their birthplaces and parents. Generally, semi-automated and human leveraging mechanisms, which can be applied to ensure the quality of knowledge graphs, are essential tools for the evaluation of knowledge graph completion. Specifically, human supervision is currently considered the gold standard evaluation in knowledge graph completion (Ballandies and Pournaras 2021 ).

Knowledge graph completion aims to expand existing knowledge graphs by adding new triplets using techniques for link prediction (Wang et al. 2020b ; Akrami et al. 2020 ) and entity prediction (Ji et al. 2021 ). These approaches typically train a machine learning model on a knowledge graph to assess the plausibility of new candidate triplets. Then, they add the candidate triplets with high plausibility to the knowledge graph. For example, for an incomplete triplet (Tom, friendOf, ?) , it is possible to assess the range of tails and return the more plausible ones to enrich the knowledge graph. These models successfully utilized knowledge graphs in many different domains, including digital libraries (Yao et al. 2017 ), biomedical (Harnoune et al. 2021 ), social media (Abu-Salih 2021 ), and scientific research (Nayyeri et al. 2021 ). Some new methods are able to process fuzzy knowledge graphs in which each triple is associated with a confidence value (Chen et al. 2019 ).

However, most current knowledge graph completion methods only focus on extracting triplets from a closed-world data source. That means the generated triplets are new, but the entities or relations in the triplets need to already exist in the knowledge graph. For example, for the incomplete triplet (Tom, friendOf, ?) , predicting the triplet (Tom, friendOf, Jerry) is only possible if the entity Jerry is already in the knowledge graph. Because of this limitation, these methods cannot add new entities and relations to the knowledge graph. To tackle this issue, we are starting to see the emergence of open-world techniques for knowledge graph completion that extracts potential objects from outside of the existing knowledge bases. For instance, the ConMask model (Shi and Weninger 2018 ) has been proposed to predict the unseen entities in knowledge graphs. However, methods for open-world knowledge graph completion still suffer from low accuracy. The main reason is that the data source is usually more complex and noisy. In addition, the similarity of the predicted new entities to the existing entities can mislead the results. In other words, two similar entities are regarded as connected entities, while they may not have a direct relationship.

Knowledge graph completion methods assume knowledge graphs are static and fail to capture the dynamic evolution of knowledge graphs. To obtain accurate facts over time, temporal knowledge graph completion, which considers the temporal information reflecting the validity of knowledge, has emerged. Compared to static knowledge graph completion, temporal knowledge graph completion methods integrate timestamps into the learning process. Hence, they explore the time-sensitive facts and improve the link prediction accuracy significantly. Although temporal knowledge graph completion methods have shown brilliant performance, they still face serious challenges. Because these models consider time information would be less efficient (Shao et al. 2022 ), the key challenge of temporal knowledge graph completion is how to effectively incorporate timestamps of facts into the learning models and properly capture the temporal dynamics of facts.

Knowledge fusion aims to combine and integrate knowledge from different data sources. It is often a necessary step for the generation of knowledge graphs (Nguyen et al. 2020 ; Smirnov and Levashova 2019 ). The primary method of knowledge fusion is entity alignment or ontology alignment (Ren et al. 2021 ), which aims to match the same entity from multiple knowledge graphs (Zhao et al. 2020 ). Achieving efficient and accurate knowledge graph fusion is a challenging task because of the complexity, variety, and large volume of data available today.

While a lot of work has been done in this direction, there are still several intriguing research directions that deserve to be investigated in the future. One of them regards cross-language knowledge fusion (Mao et al. 2020 ), which allows the integration of information from different languages. This is often used to support cross-lingual recommender systems (Javed et al. 2021 ). For example, Xu et al. ( 2019 ) adopted a graph-matching neural network to achieve cross-language entity alignment. However, the result of the cross-language knowledge fusion is still unsatisfactory because the accuracy of the matching entities from different languages is relatively low. Therefore, it remains a daunting challenge to explore cross-language knowledge fusion.

Another primary challenge regards entity disambiguation (Nguyen et al. 2020 ). As the polysemy problem of natural language, the same entity may have various expressions in different knowledge graphs. Hence, entity disambiguation is required before conducting entity alignment. Existing entity disambiguation methods mainly focus on discriminating and matching ambiguous entities based on extracting knowledge from texts containing rich contextual information (Zhu and Iglesias 2018 ). However, these methods can not precisely measure the semantic similarity of entities when the texts are short and have limited contextual information. Only a few works have focused on solving this issue. For example, Zhu and Iglesias (Zhu and Iglesias 2018 ) have proposed SCSNED for entity disambiguation. SCSNED measures semantic similarity based on both informative words of entities in knowledge graphs and contextual information in short texts. Although SCSNED alleviates the issue of limited contextual information to some extent, more effort is needed to improve the performance of entity disambiguation.

In addition, many knowledge fusion methods only focus on matching entities with the same modality and ignore multi-modal scenes in which knowledge is presented in different forms. Specifically, entity alignment considering only single-modality knowledge graph scenario has insignificant performance because it can not fully reflect the relationships of entities in the real world (Cheng et al. 2022b ). Recently, to solve this issue, some studies have proposed multi-modal knowledge fusion, which matches the same entities having different modalities and generates a multi-modal knowledge graph. For example, HMEA (Guo et al. 2021 ) aligns entities with multiple forms by mapping multi-modal representations into hyperbolic space. Although many researchers have worked on multi-modal knowledge fusion, it is still a critical task. Multi-modal knowledge fusion mainly aims to find equivalent entities by integrating their multi-modal features (Cheng et al. 2022b ). Nevertheless, how to efficiently incorporate the features having multiple modalities is still a tricky issue facing current methods.

The goal of knowledge reasoning is to infer new knowledge, such as the implicit relations between two entities (Liu et al. 2021 ; Wang et al. 2019b ), based on existing data. For a given knowledge graph, wherein there are two unconnected entities h and t , denoted as h , t ∈ G , here G means the knowledge graph, knowledge reasoning can find out the potential relation r between these entities and form a new triplet ( h ,  r ,  t ). The knowledge reasoning methods are mainly categorized into logic rule-based (De Meester et al. 2021 ), distributed representation-based (Chen et al. 2020b ), and neural network-based methods (Xiong et al. 2017 ). Logic rule-based knowledge reasoning aims to discover knowledge according to the random walk and logic rules, while distributed representation-based knowledge reasoning embeds entities and relations into a vector space to obtain distributed representation (Chen et al. 2020b ). Neural network-based knowledge reasoning method utilizes neural networks to infer new triplets given the body of knowledge in the graph (Xian et al. 2019 ).

There are two tasks in knowledge reasoning: single-hop prediction and multi-hop reasoning (Ren et al. 2022 ). Single-hop prediction predicts one element of a triplet for the given two elements, while multi-hop reasoning predicts one or more elements in a multi-hop logical query. In other words, in the multi-hop reasoning scenario, finding the answer to a typical question and forming new triplets requires the prediction and imputation of multiple edges and nodes. Multi-hop reasoning achieves a more precise formation of triplets when compared with the single-hop prediction. Therefore, multi-hop reasoning has attracted more attention and become a critical need for the development of knowledge graphs in recent years. Although many works have been done, multi-hop reasoning over knowledge graphs remains largely unexplored. Notably, multi-hop reasoning on massive knowledge graphs is one of the challenging tasks (Zhu et al. 2022 ). For instance, most recent studies focus on multi-hop reasoning over knowledge graphs, which have only 63K entities and 592K relations. The existing models can’t learn the training set effectively for a massive knowledge graph that has more than millions of entities. Moreover, multi-hop reasoning needs to traverse multiple relations and intermediate entities in the knowledge graph, which could lead to exponential computation cost (Zhang et al. 2021 ). Therefore, it is still a daunting task to explore multi-hop knowledge reasoning.

Besides, the verification of inferred new knowledge is also a critical issue. Knowledge reasoning enriches existing knowledge graphs and brings benefits to the downstream tasks (Wan et al. 2021 ). However, the inferred new knowledge is sometimes uncertain, and the veracity of new triplets needs to be verified. Furthermore, the conflicts between new and existing knowledge should be detected. To address these problems, some research has proposed multi-source knowledge reasoning (Zhao et al. 2020 ) that detects erroneous knowledge and conflicting knowledge. Overall, more attention should be paid to multi-source knowledge reasoning and erroneous knowledge reduction.

Conclusions

Knowledge graphs have played an instrumental role in creating many intelligent services and applications for various fields. In this survey, we provided an overview of knowledge graphs in terms of opportunities and challenges. We first introduced the definitions and existing research directions regarding knowledge graphs to provide an introductory analysis of knowledge graphs. Afterward, we discussed AI systems that take advantage of knowledge graphs. Then, we presented some representative knowledge graph applications in several fields. Furthermore, we analyzed the limitations of current knowledge graph technologies, which lead to severe technical challenges. We expect this survey to spark new ideas and insightful perspectives for future research and development activities involving knowledge graphs.

Open Access funding enabled and organized by CAUL and its Member Institutions.

Declarations

The authors declare that they have no competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

1 AMiner - https://www.aminer.cn/

2 Open Academic Graph - https://www.openacademic.ai/oag/

3 AI-KG - https://w3id.org/aikg/

4 AIDA - http://w3id.org/aida

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Contributor Information

Ciyuan Peng, Email: [email protected] .

Feng Xia, Email: [email protected] .

Mehdi Naseriparsa, Email: [email protected] .

Francesco Osborne, Email: [email protected] .

• Abu-Salih B. Domain-specific knowledge graphs: a survey. J Netw Comput Appl. 2021; 185 (103):076. [ Google Scholar ]
• Akrami F, Saeef MS, Zhang Q et al (2020) Realistic re-evaluation of knowledge graph completion methods: an experimental study. In: Proceedings of the 2020 ACM SIGMOD International Conference on Management of Data, pp 1995–2010
• Aliyu I, Kana A, Aliyu S. Development of knowledge graph for university courses management. Int J Educ Manag Eng. 2020; 10 (2):1. [ Google Scholar ]
• An B, Chen B, Han X et al (2018) Accurate text-enhanced knowledge graph representation learning. In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, vol. 1 (Long Papers), pp 745–755
• Angioni S, Salatino A, Osborne F et al (2021) Aida: a knowledge graph about research dynamics in academia and industry. Quant Sci Stud p 1–43
• Auer S, Bizer C, Kobilarov G et al (2007) Dbpedia: a nucleus for a web of open data. In: The semantic web. Springer, p 722–735
• Bai X, Wang M, Lee I, et al. Scientific paper recommendation: a survey. IEEE Access. 2019; 7 :9324–9339. doi: 10.1109/ACCESS.2018.2890388. [ CrossRef ] [ Google Scholar ]
• Bai X, Zhang F, Li J et al (2021) Educational big data: prediction, applications and challenges. Big Data Res 26(100270)
• Baken N (2020) Linked data for smart homes: comparing rdf and labeled property graphs. In: LDAC2020–8th linked data in architecture and construction workshop, p 23–36
• Balažević I, Allen C, Hospedales TM (2019) Tucker: tensor factorization for knowledge graph completion. arXiv preprint arXiv:1901.09590
• Ballandies MC, Pournaras E. Mobile link prediction: automated creation and crowdsourced validation of knowledge graphs. Microprocess Microsyst. 2021; 87 (104):335. [ Google Scholar ]
• Bauer L, Wang Y, Bansal M (2018) Commonsense for generative multi-hop question answering tasks. arXiv preprint arXiv:1809.06309
• Bekoulis G, Deleu J, Demeester T, et al. Joint entity recognition and relation extraction as a multi-head selection problem. Expert Syst Appl. 2018; 114 :34–45. doi: 10.1016/j.eswa.2018.07.032. [ CrossRef ] [ Google Scholar ]
• Bollacker K, Evans C, Paritosh P et al (2008) Freebase: a collaboratively created graph database for structuring human knowledge. In: Proceedings of the 2008 ACM SIGMOD international conference on management of data, p 1247–1250
• Bordes A, Glorot X, Weston J, et al. A semantic matching energy function for learning with multi-relational data. Mach Learn. 2014; 94 (2):233–259. doi: 10.1007/s10994-013-5363-6. [ CrossRef ] [ Google Scholar ]
• Bordes A, Usunier N, Garcia-Duran A et al (2013) Translating embeddings for modeling multi-relational data. Adv Neural Inf Process Syst 26
• Bordes A, Weston J, Collobert R et al (2011) Learning structured embeddings of knowledge bases. In: Twenty-fifth AAAI conference on artificial intelligence
• Bounhas I, Soudani N, Slimani Y. Building a morpho-semantic knowledge graph for Arabic information retrieval. Info Process Manag. 2020; 57 (6):102. [ Google Scholar ]
• Cai L, Wang WY (2017) Kbgan: adversarial learning for knowledge graph embeddings. arXiv preprint arXiv:1711.04071
• Chaudhri V, Baru C, Chittar N, et al. Knowledge graphs: introduction, history and perspectives. AI Mag. 2022; 43 (1):17–29. [ Google Scholar ]
• Chen P, Lu Y, Zheng VW, et al. Knowedu: a system to construct knowledge graph for education. IEEE Access. 2018; 6 :31553–31563. doi: 10.1109/ACCESS.2018.2839607. [ CrossRef ] [ Google Scholar ]
• Chen R, Chen T, Hui X et al (2020a) Knowledge graph transfer network for few-shot recognition. In: Proceedings of the AAAI conference on artificial intelligence, p 10,575–10,582
• Chen X, Jia S, Xiang Y. A review: knowledge reasoning over knowledge graph. Expert Syst Appl. 2020; 141 (112):948. [ Google Scholar ]
• Chen YC, Hui L, Thaipisutikul T et al (2020c) A collaborative filtering recommendation system with dynamic time decay. J Supercomput p 1–19
• Chen X, Chen M, Shi W et al (2019) Embedding uncertain knowledge graphs. In: Proceedings of the AAAI Conference on Artificial Intelligence, p 3363–3370
• Cheng D, Yang F, Xiang S, et al. Financial time series forecasting with multi-modality graph neural network. Pattern Recogn. 2022; 121 (108):218. [ Google Scholar ]
• Cheng B, Zhu J, Guo M (2022b) Multijaf: multi-modal joint entity alignment framework for multi-modal knowledge graph. Neurocomputing
• Chi Y, Qin Y, Song R, et al. Knowledge graph in smart education: a case study of entrepreneurship scientific publication management. Sustainability. 2018; 10 (4):995. doi: 10.3390/su10040995. [ CrossRef ] [ Google Scholar ]
• Choi D, Chun S, Oh H, et al. Rumor propagation is amplified by echo chambers in social media. Sci Rep. 2020; 10 (1):1–10. [ PMC free article ] [ PubMed ] [ Google Scholar ]
• Cui L, Seo H, Tabar M et al (2020) Deterrent: knowledge guided graph attention network for detecting healthcare misinformation. In: Proceedings of the 26th ACM SIGKDD international conference on knowledge discovery & data mining, p 492–502
• Dai Y, Wang S, Chen X, et al. Generative adversarial networks based on Wasserstein distance for knowledge graph embeddings. Knowl-Based Syst. 2020; 190 (105):165. [ Google Scholar ]
• Dai Y, Wang S, Xiong NN, et al. A survey on knowledge graph embedding: approaches, applications and benchmarks. Electronics. 2020; 9 (5):750. doi: 10.3390/electronics9050750. [ CrossRef ] [ Google Scholar ]
• Dalton J, Dietz L, Allan J (2014) Entity query feature expansion using knowledge base links. In: Proceedings of the 37th international ACM SIGIR conference on research & development in information retrieval, p 365–374
• d’Aquin M (2016) On the use of linked open data in education: current and future practices. In: Open data for education. Springer, p 3–15
• Das A, Mandal J, Danial Z, et al. An improvement of Bengali factoid question answering system using unsupervised statistical methods. Sādhanā 2022; 47 (1):1–14. doi: 10.1007/s12046-021-01765-3. [ CrossRef ] [ Google Scholar ]
• De Meester B, Heyvaert P, Arndt D et al (2021) Rdf graph validation using rule-based reasoning. Semantic Web (Preprint):1–26
• Dessì D, Osborne F, Recupero DR et al (2020) AI-KG: an automatically generated knowledge graph of artificial intelligence. In: ISWC 2020, vol 12507. Springer, p 127–143
• Dimitrakis E, Sgontzos K, Tzitzikas Y. A survey on question answering systems over linked data and documents. J Intell Inf Syst. 2020; 55 (2):233–259. doi: 10.1007/s10844-019-00584-7. [ CrossRef ] [ Google Scholar ]
• Ehrlinger L, Wöß W. Towards a definition of knowledge graphs. SEMANTiCS (Posters, Demos, SuCCESS) 2016; 48 (1–4):2. [ Google Scholar ]
• Fan W, Ma Y, Li Q et al (2019) Graph neural networks for social recommendation. In: The world wide web conference, p 417–426
• Färber M, Bartscherer F, Menne C, et al. Linked data quality of dbpedia, freebase, opencyc, wikidata, and yago. Semantic Web. 2018; 9 (1):77–129. doi: 10.3233/SW-170275. [ CrossRef ] [ Google Scholar ]
• Farfán F, Hristidis V, Ranganathan A et al (2009) Xontorank: Ontology-aware search of electronic medical records. In: Proceedings of the 25th International Conference on Data Engineering, ICDE 2009, March 29 2009–April 2 2009, Shanghai, China. IEEE Computer Society, p 820–831
• Fu TJ, Li PH, Ma WY (2019) Graphrel: modeling text as relational graphs for joint entity and relation extraction. In: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, p 1409–1418
• Gao Y, Li YF, Lin Y et al (2020) Deep learning on knowledge graph for recommender system: a survey. arXiv preprint arXiv:2004.00387
• Gómez E, Zhang CS, Boratto L, et al. Enabling cross-continent provider fairness in educational recommender systems. Futur Gener Comput Syst. 2022; 127 :435–447. doi: 10.1016/j.future.2021.08.025. [ CrossRef ] [ Google Scholar ]
• Gong F, Wang M, Wang H, et al. Smr: medical knowledge graph embedding for safe medicine recommendation. Big Data Res. 2021; 23 (100):174. [ Google Scholar ]
• Guo H, Tang J, Zeng W, et al. Multi-modal entity alignment in hyperbolic space. Neurocomputing. 2021; 461 :598–607. doi: 10.1016/j.neucom.2021.03.132. [ CrossRef ] [ Google Scholar ]
• Guo S, Wang Q, Wang B et al (2015) Semantically smooth knowledge graph embedding. In: Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (vol 1: Long Papers), p 84–94
• Guo Q, Zhuang F, Qin C et al (2020) A survey on knowledge graph-based recommender systems. IEEE Trans Knowl Data Eng
• Harnoune A, Rhanoui M, Mikram M, et al. Bert based clinical knowledge extraction for biomedical knowledge graph construction and analysis. Comput Methods Programs Biomed Update. 2021; 1 (100):042. [ Google Scholar ]
• Hashemi M, Hall M. Multi-label classification and knowledge extraction from oncology-related content on online social networks. Artif Intell Rev. 2020; 53 (8):5957–5994. doi: 10.1007/s10462-020-09839-0. [ CrossRef ] [ Google Scholar ]
• He S, Liu K, Ji G et al (2015) Learning to represent knowledge graphs with gaussian embedding. In: Proceedings of the 24th ACM international on conference on information and knowledge management, p 623–632
• Hersh W (2021) Information retrieval. In: Biomedical informatics. Springer, p 755–794
• Hogan A, Blomqvist E, Cochez M, et al. Knowledge graphs. ACM Comput Surveys (CSUR) 2021; 54 (4):1–37. doi: 10.1145/3447772. [ CrossRef ] [ Google Scholar ]
• Huang X, Zhang J, Li D et al (2019) Knowledge graph embedding based question answering. In: Proceedings of the Twelfth ACM International Conference on Web Search and Data Mining, p 105–113
• Javed U, Shaukat K, Hameed IA, et al. A review of content-based and context-based recommendation systems. Int J Emerg Technol Learning. 2021; 16 (3):274–306. doi: 10.3991/ijet.v16i03.18851. [ CrossRef ] [ Google Scholar ]
• Ji G, He S, Xu L et al (2015) Knowledge graph embedding via dynamic mapping matrix. In: Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (vol 1: Long Papers), p 687–696
• Ji G, Liu K, He S et al (2016) Knowledge graph completion with adaptive sparse transfer matrix. In: Thirtieth AAAI conference on artificial intelligence
• Ji S, Pan S, Cambria E et al (2021) A survey on knowledge graphs: representation, acquisition, and applications. IEEE Trans Neural Netw Learn Syst [ PubMed ]
• Jia Y, Wang Y, Lin H et al (2016) Locally adaptive translation for knowledge graph embedding. In: Thirtieth AAAI conference on artificial intelligence
• Katzman JL, Shaham U, Cloninger A, et al. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural network. BMC Med Res Methodol. 2018; 18 (1):1–12. doi: 10.1186/s12874-018-0482-1. [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Kazemi SM, Poole D (2018) Simple embedding for link prediction in knowledge graphs. Adv Neural Inf Process Syst 31
• Ko H, Witherell P, Lu Y, et al. Machine learning and knowledge graph based design rule construction for additive manufacturing. Addit Manuf. 2021; 37 (101):620. [ Google Scholar ]
• Kong Y, Liu X, Zhao Z, et al. Bolt defect classification algorithm based on knowledge graph and feature fusion. Energy Rep. 2022; 8 :856–863. doi: 10.1016/j.egyr.2021.11.127. [ CrossRef ] [ Google Scholar ]
• Li J, Cai T, Deng K, et al. Community-diversified influence maximization in social networks. Inf Syst. 2020; 92 (101):522. [ Google Scholar ]
• Li L, Wang P, Yan J, et al. Real-world data medical knowledge graph: construction and applications. Artif Intell Med. 2020; 103 (101):817. [ PubMed ] [ Google Scholar ]
• Li Z, Liu H, Zhang Z et al (2021) Learning knowledge graph embedding with heterogeneous relation attention networks. IEEE Trans Neural Netw Learn Syst [ PubMed ]
• Liang B, Su H, Gui L, et al. Aspect-based sentiment analysis via affective knowledge enhanced graph convolutional networks. Knowl-Based Syst. 2022; 235 (107):643. [ Google Scholar ]
• Lin Y, Liu Z, Sun M et al (2015) Learning entity and relation embeddings for knowledge graph completion. In: Twenty-ninth AAAI conference on artificial intelligence
• Lin X, Quan Z, Wang ZJ et al (2020) Kgnn: Knowledge graph neural network for drug-drug interaction prediction. In: IJCAI, p 2739–2745
• Liu J, Kong X, Zhou X, et al. Data mining and information retrieval in the 21st century: a bibliographic review. Comput Sci Rev. 2019; 34 (100):193. [ Google Scholar ]
• Liu J, Xia F, Wang L, et al. Shifu2: a network representation learning based model for advisor-advisee relationship mining. IEEE Trans Knowl Data Eng. 2021; 33 (4):1763–1777. [ Google Scholar ]
• Liu J, Ren J, Zheng W et al (2020) Web of scholars: A scholar knowledge graph. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pp 2153–2156
• Liu Q, Jiang H, Evdokimov A et al (2016) Probabilistic reasoning via deep learning: Neural association models. arXiv preprint arXiv:1603.07704
• Liu Z, Xiong C, Sun M et al (2018) Entity-duet neural ranking: Understanding the role of knowledge graph semantics in neural information retrieval. arXiv preprint arXiv:1805.07591
• MacLean F. Knowledge graphs and their applications in drug discovery. Expert Opin Drug Discov. 2021; 16 (9):1057–1069. doi: 10.1080/17460441.2021.1910673. [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Mao X, Wang W, Xu H et al (2020) Mraea: an efficient and robust entity alignment approach for cross-lingual knowledge graph. In: Proceedings of the 13th International Conference on Web Search and Data Mining, p 420–428
• Mayank M, Sharma S, Sharma R (2021) Deap-faked: knowledge graph based approach for fake news detection. arXiv preprint arXiv:2107.10648
• Meel P, Vishwakarma DK. Fake news, rumor, information pollution in social media and web: a contemporary survey of state-of-the-arts, challenges and opportunities. Expert Syst Appl. 2020; 153 (112):986. [ Google Scholar ]
• Messner J, Abboud R, Ceylan II (2022) Temporal knowledge graph completion using box embeddings. In: Proceedings of the AAAI Conference on Artificial Intelligence, pp 7779–7787
• Mikolov T, Chen K, Corrado G et al (2013) Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781
• Minervini P, Bošnjak M, Rocktäschel T et al (2020) Differentiable reasoning on large knowledge bases and natural language. In: Proceedings of the AAAI conference on artificial intelligence, p 5182–5190
• Mohamed SK, Nounu A, Nováček V. Biological applications of knowledge graph embedding models. Brief Bioinform. 2021; 22 (2):1679–1693. doi: 10.1093/bib/bbaa012. [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Monti D, Rizzo G, Morisio M. A systematic literature review of multicriteria recommender systems. Artif Intell Rev. 2021; 54 :427–468. doi: 10.1007/s10462-020-09851-4. [ CrossRef ] [ Google Scholar ]
• Naseriparsa M, Islam MS, Liu C, et al. No-but-semantic-match: computing semantically matched xml keyword search results. World Wide Web. 2018; 21 (5):1223–1257. doi: 10.1007/s11280-017-0503-8. [ CrossRef ] [ Google Scholar ]
• Naseriparsa M, Liu C, Islam MS, et al. Xplorerank: exploring XML data via you may also like queries. World Wide Web. 2019; 22 (4):1727–1750. doi: 10.1007/s11280-018-0630-x. [ CrossRef ] [ Google Scholar ]
• Naseriparsa M, Islam MS, Liu C et al (2019b) Xsnippets: exploring semi-structured data via snippets. Data Knowl Eng 124
• Nayyeri M, Cil GM, Vahdati S, et al. Trans4e: link prediction on scholarly knowledge graphs. Neurocomputing. 2021; 461 :530–542. doi: 10.1016/j.neucom.2021.02.100. [ CrossRef ] [ Google Scholar ]
• Nguyen DQ, Nguyen TD, Nguyen DQ et al (2017) A novel embedding model for knowledge base completion based on convolutional neural network. In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, p 327–333
• Nguyen DQ, Sirts K, Qu L et al (2016) Stranse: a novel embedding model of entities and relationships in knowledge bases. arXiv preprint arXiv:1606.08140
• Nguyen HL, Vu DT, Jung JJ. Knowledge graph fusion for smart systems: a survey. Info Fusion. 2020; 61 :56–70. doi: 10.1016/j.inffus.2020.03.014. [ CrossRef ] [ Google Scholar ]
• Nickel M, Rosasco L, Poggio T (2016) Holographic embeddings of knowledge graphs. In: Proceedings of the AAAI Conference on Artificial Intelligence
• Nickel M, Tresp V, Kriegel HP (2011) A three-way model for collective learning on multi-relational data. In: ICML
• Noy N, Gao Y, Jain A, et al. Industry-scale knowledge graphs: lessons and challenges: five diverse technology companies show how it’s done. Queue. 2019; 17 (2):48–75. doi: 10.1145/3329781.3332266. [ CrossRef ] [ Google Scholar ]
• Palumbo E, Monti D, Rizzo G, et al. entity2rec: property-specific knowledge graph embeddings for item recommendation. Expert Syst Appl. 2020; 151 (113):235. [ Google Scholar ]
• Palumbo E, Rizzo G, Troncy R et al (2018) Knowledge graph embeddings with node2vec for item recommendation. In: European Semantic Web Conference, Springer, p 117–120
• Pedersen T, Patwardhan S, Michelizzi J et al (2004) Wordnet: similarity-measuring the relatedness of concepts. In: AAAI, p 25–29
• Peng C, Vu DT, Jung JJ (2021) Knowledge graph-based metaphor representation for literature understanding. Digital Scholarship Humanities
• Pereira CK, Siqueira SWM, Nunes BP, et al. Linked data in education: a survey and a synthesis of actual research and future challenges. IEEE Trans Learn Technol. 2017; 11 (3):400–412. doi: 10.1109/TLT.2017.2787659. [ CrossRef ] [ Google Scholar ]
• Qiu Y, Wang Y, Jin X et al (2020) Stepwise reasoning for multi-relation question answering over knowledge graph with weak supervision. In: Proceedings of the 13th International Conference on Web Search and Data Mining, p 474–482
• Quijano-Sánchez L, Cantador I, Cortés-Cediel ME, et al. Recommender systems for smart cities. Inf Syst. 2020; 92 (101):545. [ Google Scholar ]
• Rebele T, Suchanek F, Hoffart J et al (2016) Yago: a multilingual knowledge base from wikipedia, wordnet, and geonames. In: International semantic web conference, Springer, p 177–185
• Ren J, Xia F, Chen X, et al. Matching algorithms: fundamentals, applications and challenges. IEEE Trans Emerg Top Comput Intell. 2021; 5 (3):332–350. doi: 10.1109/TETCI.2021.3067655. [ CrossRef ] [ Google Scholar ]
• Ren H, Dai H, Dai B et al (2022) Smore: Knowledge graph completion and multi-hop reasoning in massive knowledge graphs. In: Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, p 1472–1482
• Rodriguez-Muro M, Rezk M. Efficient sparql-to-sql with r2rml mappings. J Web Semantics. 2015; 33 :141–169. doi: 10.1016/j.websem.2015.03.001. [ CrossRef ] [ Google Scholar ]
• Rossi A, Barbosa D, Firmani D, et al. Knowledge graph embedding for link prediction: a comparative analysis. ACM Trans Knowl Discov Data (TKDD) 2021; 15 (2):1–49. doi: 10.1145/3424672. [ CrossRef ] [ Google Scholar ]
• Salatino AA, Thanapalasingam T, Mannocci A et al (2020) The computer science ontology: a comprehensive automatically-generated taxonomy of research areas. Data Intell 2(3)
• Saraji MK, Mardani A, Köppen M, et al. An extended hesitant fuzzy set using swara-multimoora approach to adapt online education for the control of the pandemic spread of covid-19 in higher education institutions. Artif Intell Rev. 2022; 55 (1):181–206. doi: 10.1007/s10462-021-10029-9. [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Saxena A, Tripathi A, Talukdar P (2020) Improving multi-hop question answering over knowledge graphs using knowledge base embeddings. In: Proceedings of the 58th annual meeting of the association for computational linguistics, p 4498–4507
• Schlichtkrull M, Kipf TN, Bloem P et al (2018) Modeling relational data with graph convolutional networks. In: European semantic web conference, Springer, p 593–607
• Shao B, Li X, Bian G. A survey of research hotspots and frontier trends of recommendation systems from the perspective of knowledge graph. Expert Syst Appl. 2021; 165 (113):764. [ Google Scholar ]
• Shao P, Zhang D, Yang G, et al. Tucker decomposition-based temporal knowledge graph completion. Knowl-Based Syst. 2022; 238 (107):841. [ Google Scholar ]
• Shi B, Weninger T (2018) Open-world knowledge graph completion. In: Thirty-Second AAAI Conference on Artificial Intelligence
• Shi C, Ding J, Cao X, et al. Entity set expansion in knowledge graph: a heterogeneous information network perspective. Front Comp Sci. 2021; 15 (1):1–12. [ Google Scholar ]
• Shin S, Jin X, Jung J, et al. Predicate constraints based question answering over knowledge graph. Info Process Manag. 2019; 56 (3):445–462. doi: 10.1016/j.ipm.2018.12.003. [ CrossRef ] [ Google Scholar ]
• Shokeen J, Rana C. A study on features of social recommender systems. Artif Intell Rev. 2020; 53 (2):965–988. doi: 10.1007/s10462-019-09684-w. [ CrossRef ] [ Google Scholar ]
• Shu H, Huang J (2021) User-preference based knowledge graph feature and structure learning for recommendation. In: 2021 IEEE International Conference on Multimedia and Expo (ICME), IEEE, p 1–6
• Singh K, Lytra I, Radhakrishna AS, et al. No one is perfect: analysing the performance of question answering components over the dbpedia knowledge graph. J Web Semantics. 2020; 65 (100):594. [ Google Scholar ]
• Smirnov A, Levashova T. Knowledge fusion patterns: a survey. Inf Fusion. 2019; 52 :31–40. doi: 10.1016/j.inffus.2018.11.007. [ CrossRef ] [ Google Scholar ]
• Socher R, Chen D, Manning CD et al (2013) Reasoning with neural tensor networks for knowledge base completion. In: Advances in neural information processing systems, p 926–934
• Sun J, Xu J, Zheng K et al (2017) Interactive spatial keyword querying with semantics. In: Proceedings of the 2017 ACM on Conference on Information and Knowledge Management, CIKM 2017, Singapore, November 06–10, 2017. ACM, p 1727–1736
• Sun K, Yu S, Peng C, et al. Relational structure-aware knowledge graph representation in complex space. Mathematics. 2022; 10 (11):1930. doi: 10.3390/math10111930. [ CrossRef ] [ Google Scholar ]
• Sun R, Cao X, Zhao Y et al (2020) Multi-modal knowledge graphs for recommender systems. In: Proceedings of the 29th ACM International Conference on Information & Knowledge Management, p 1405–1414
• Sun Z, Deng ZH, Nie JY et al (2019a) Rotate: knowledge graph embedding by relational rotation in complex space. arXiv preprint arXiv:1902.10197
• Sun Z, Guo Q, Yang J, et al. Research commentary on recommendations with side information: a survey and research directions. Electron Commer Res Appl. 2019; 37 (100):879. [ Google Scholar ]
• Trouillon T, Welbl J, Riedel S et al (2016) Complex embeddings for simple link prediction. In: International conference on machine learning, PMLR, p 2071–2080
• Ugander J, Karrer B, Backstrom L et al (2011) The anatomy of the facebook social graph. arXiv preprint arXiv:1111.4503
• Vashishth S, Sanyal S, Nitin V et al (2020) Interacte: improving convolution-based knowledge graph embeddings by increasing feature interactions. In: Proceedings of the AAAI Conference on Artificial Intelligence, p 3009–3016
• Vrandečić D, Krötzsch M. Wikidata: a free collaborative knowledgebase. Commun ACM. 2014; 57 (10):78–85. doi: 10.1145/2629489. [ CrossRef ] [ Google Scholar ]
• Wan L, Xia F, Kong X, et al. Deep matrix factorization for trust-aware recommendation in social networks. IEEE Trans Netw Sci Eng. 2020; 8 (1):511–528. doi: 10.1109/TNSE.2020.3044035. [ CrossRef ] [ Google Scholar ]
• Wang C, Yu H, Wan F (2018a) Information retrieval technology based on knowledge graph. In: 2018 3rd International Conference on Advances in Materials, Mechatronics and Civil Engineering (ICAMMCE 2018), Atlantis Press, p 291–296
• Wang H, Zhang F, Wang J et al (2018b) Ripplenet: Propagating user preferences on the knowledge graph for recommender systems. In: Proceedings of the 27th ACM International Conference on Information and Knowledge Management, p 417–426
• Wang R, Yan Y, Wang J et al (2018c) Acekg: a large-scale knowledge graph for academic data mining. In: Proceedings of the 27th ACM International Conference on Information and Knowledge Management. Association for Computing Machinery, New York, NY, CIKM ’18, p 1487–1490
• Wang Z, Chen T, Ren J et al (2018d) Deep reasoning with knowledge graph for social relationship understanding. arXiv preprint arXiv:1807.00504
• Wang K, Shen Z, Huang C, et al. Microsoft academic graph: when experts are not enough. Quant Sci Stud. 2020; 1 (1):396–413. doi: 10.1162/qss_a_00021. [ CrossRef ] [ Google Scholar ]
• Wang L, Ren J, Xu B, et al. Model: motif-based deep feature learning for link prediction. IEEE Trans Comput Soc Syst. 2020; 7 (2):503–516. doi: 10.1109/TCSS.2019.2962819. [ CrossRef ] [ Google Scholar ]
• Wang W, Liu J, Tang T, et al. Attributed collaboration network embedding for academic relationship mining. ACM Trans Web (TWEB) 2020; 15 (1):1–20. [ Google Scholar ]
• Wang Z, Yin Z, Argyris YA. Detecting medical misinformation on social media using multimodal deep learning. IEEE J Biomed Health Info. 2020; 25 (6):2193–2203. doi: 10.1109/JBHI.2020.3037027. [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Wang Q, Li M, Wang X et al (2020e) Covid-19 literature knowledge graph construction and drug repurposing report generation. arXiv preprint arXiv:2007.00576
• Wang W, Liu J, Yang Z, et al. Sustainable collaborator recommendation based on conference closure. IEEE Trans Comput Soc Syst. 2019; 6 (2):311–322. doi: 10.1109/TCSS.2019.2898198. [ CrossRef ] [ Google Scholar ]
• Wang X, Wang D, Xu C et al (2019b) Explainable reasoning over knowledge graphs for recommendation. In: Proceedings of the AAAI Conference on Artificial Intelligence, p 5329–5336
• Wang H, Zhang F, Zhao M et al (2019c) Multi-task feature learning for knowledge graph enhanced recommendation. In: The World Wide Web Conference, p 2000–2010
• Wang Y, Dong L, Li Y, et al. Multitask feature learning approach for knowledge graph enhanced recommendations with Ripplenet. Plos One. 2021; 16 (5):e0251. doi: 10.1371/journal.pone.0251162. [ PMC free article ] [ PubMed ] [ CrossRef ] [ Google Scholar ]
• Wang Z, Zhang J, Feng J et al (2014) Knowledge graph embedding by translating on hyperplanes. In: Proceedings of the AAAI Conference on Artificial Intelligence
• Wan G, Pan S, Gong C et al (2021) Reasoning like human: hierarchical reinforcement learning for knowledge graph reasoning. In: Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, p 1926–1932
• Wise C, Ioannidis VN, Calvo MR et al (2020) Covid-19 knowledge graph: accelerating information retrieval and discovery for scientific literature. arXiv preprint arXiv:2007.12731
• Wu Y, Yang S, Yan X (2013) Ontology-based subgraph querying. In: 29th IEEE International Conference on Data Engineering, ICDE 2013, Brisbane, Australia, April 8-12, 2013. IEEE Computer Society, p 697–708
• Xia F, Asabere NY, Liu H, et al. Socially aware conference participant recommendation with personality traits. IEEE Syst J. 2014; 11 (4):2255–2266. doi: 10.1109/JSYST.2014.2342375. [ CrossRef ] [ Google Scholar ]
• Xia F, Liu H, Asabere NY et al (2014b) Multi-category item recommendation using neighborhood associations in trust networks. In: Proceedings of the 23rd International Conference on World Wide Web, p 403–404
• Xia F, Liu H, Lee I, et al. Scientific article recommendation: exploiting common author relations and historical preferences. IEEE Trans Big Data. 2016; 2 (2):101–112. doi: 10.1109/TBDATA.2016.2555318. [ CrossRef ] [ Google Scholar ]
• Xia F, Sun K, Yu S, et al. Graph learning: a survey. IEEE Trans Artif Intell. 2021; 2 (2):109–127. doi: 10.1109/TAI.2021.3076021. [ CrossRef ] [ Google Scholar ]
• Xian Y, Fu Z, Muthukrishnan S et al (2019) Reinforcement knowledge graph reasoning for explainable recommendation. In: Proceedings of the 42nd international ACM SIGIR conference on research and development in information retrieval, p 285–294
• Xiao H, Huang M, Hao Y et al (2015) Transg: a generative mixture model for knowledge graph embedding. arXiv preprint arXiv:1509.05488
• Xiong W, Hoang T, Wang WY (2017) Deep path: a reinforcement learning method for knowledge graph reasoning. arXiv preprint arXiv:1707.06690
• Xu J, Yu S, Sun K, et al. Multivariate relations aggregation learning in social networks. Proc ACM/IEEE Joint Conf Digital Libraries in. 2020; 2020 :77–86. [ Google Scholar ]
• Xu K, Wang L, Yu M et al (2019) Cross-lingual knowledge graph alignment via graph matching neural network. arXiv preprint arXiv:1905.11605
• Yao L, Mao C, Luo Y (2019) Kg-bert: Bert for knowledge graph completion. arXiv preprint arXiv:1909.03193
• Yao L, Zhang Y, Wei B et al (2017) Incorporating knowledge graph embeddings into topic modeling. In: Thirty-First AAAI Conference on Artificial Intelligence
• Yao S, Wang R, Sun S et al (2020) Joint embedding learning of educational knowledge graphs. In: Artificial Intelligence Supported Educational Technologies p 209–224
• Ying R, He R, Chen K et al (2018) Graph convolutional neural networks for web-scale recommender systems. In: Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining, p 974–983
• Yong Y, Yao Z, Zhao Y (2021) A framework for reviewer recommendation based on knowledge graph and rules matching. In: 2021 IEEE International Conference on Information Communication and Software Engineering (ICICSE), p 199–203
• Yu H, Li H, Mao D, et al. A relationship extraction method for domain knowledge graph construction. World Wide Web. 2020; 23 (2):735–753. doi: 10.1007/s11280-019-00765-y. [ CrossRef ] [ Google Scholar ]
• Yuan H, Deng W (2021) Doctor recommendation on healthcare consultation platforms: an integrated framework of knowledge graph and deep learning. Internet Research
• Zablith F (2022) Constructing social media links to formal learning: a knowledge graph approach. Educational technology research and development p 1–26
• Zhang H, Fang Q, Qian S et al (2019a) Multi-modal knowledge-aware event memory network for social media rumor detection. In: Proceedings of the 27th ACM International Conference on Multimedia, p 1942–1951
• Zhang N, Deng S, Sun Z et al (2019b) Long-tail relation extraction via knowledge graph embeddings and graph convolution networks. arXiv preprint arXiv:1903.01306
• Zhang S, Tay Y, Yao L et al (2019c) Quaternion knowledge graph embeddings. Adv Neural Info Process Syst 32
• Zhang Y, Sheng M, Zhou R, et al. Hkgb: an inclusive, extensible, intelligent, semi-auto-constructed knowledge graph framework for healthcare with clinicians’ expertise incorporated. Info Process Manag. 2020; 57 (6):102. [ Google Scholar ]
• Zhang Z, Cai J, Zhang Y et al (2020b) Learning hierarchy-aware knowledge graph embeddings for link prediction. In: Proceedings of the AAAI Conference on Artificial Intelligence, p 3065–3072
• Zhang Y, Zhang F, Yao P et al (2018) Name disambiguation in aminer: clustering, maintenance, and human in the loop. In: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, p 1002–1011
• Zhang Z, Wang J, Chen J, et al. Cone: cone embeddings for multi-hop reasoning over knowledge graphs. Adv Neural Info Process Syst. 2021; 34 :19,172–19,183. [ Google Scholar ]
• Zhao X, Jia Y, Li A, et al. Multi-source knowledge fusion: a survey. World Wide Web. 2020; 23 (4):2567–2592. doi: 10.1007/s11280-020-00811-0. [ CrossRef ] [ Google Scholar ]
• Zheng D, Song X, Ma C et al (2020) Dgl-ke: training knowledge graph embeddings at scale. In: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, p 739–748
• Zheng Y, Wang DX. A survey of recommender systems with multi-objective optimization. Neurocomputing. 2022; 474 :141–153. doi: 10.1016/j.neucom.2021.11.041. [ CrossRef ] [ Google Scholar ]
• Zhou D, Zhou B, Zheng Z et al (2022) Schere: Schema reshaping for enhancing knowledge graph construction. In: Proceedings of the 31st ACM International Conference on Information & Knowledge Management, p 5074–5078
• Zhu A, Ouyang D, Liang S, et al. Step by step: a hierarchical framework for multi-hop knowledge graph reasoning with reinforcement learning. Knowl-Based Syst. 2022; 248 (108):843. [ Google Scholar ]
• Zhu G, Iglesias CA. Exploiting semantic similarity for named entity disambiguation in knowledge graphs. Expert Syst Appl. 2018; 101 :8–24. doi: 10.1016/j.eswa.2018.02.011. [ CrossRef ] [ Google Scholar ]
• Zhu X, Li Z, Wang X et al (2022b) Multi-modal knowledge graph construction and application: a survey. arXiv preprint arXiv:2202.05786
• Zou X. A survey on application of knowledge graph. J Phys Conf Ser. 2020; 1487 (012):016. [ Google Scholar ]