speech on school violence

• Bipolar Disorder
• Therapy Center
• When To See a Therapist
• Types of Therapy
• Best Online Therapy
• Best Couples Therapy
• Best Family Therapy
• Managing Stress
• Sleep and Dreaming
• Understanding Emotions
• Self-Improvement
• Healthy Relationships
• Student Resources
• Personality Types
• Verywell Mind Insights
• 2023 Verywell Mind 25
• Mental Health in the Classroom
• Editorial Process
• Meet Our Review Board
• Crisis Support

How to Identify and Prevent School Violence

Sanjana is a health writer and editor. Her work spans various health-related topics, including mental health, fitness, nutrition, and wellness.

Ann-Louise T. Lockhart, PsyD, ABPP, is a board-certified pediatric psychologist, parent coach, author, speaker, and owner of A New Day Pediatric Psychology, PLLC.

Yasser Chalid / Getty Images

Recognizing the Signs of School Violence

School violence refers to violence that takes place in a school setting. This includes violence on school property, on the way to or from school, and at school trips and events. It may be committed by students, teachers, or other members of the school staff; however, violence by fellow students is the most common.

An estimated 246 million children experience school violence every year; however, girls and gender non-conforming people are disproportionately affected.

"School violence can be anything that involves a real or implied threat—it can be verbal, sexual, or physical, and perpetrated with or without weapons. If someone is deliberately harming someone or acting in a way that leaves someone feeling threatened, that‘s school violence,” says Aimee Daramus , PsyD, a licensed clinical psychologist.

This article explores the types, causes, and impact of school violence and suggests some steps that can help prevent it.

Types of School Violence

School violence can take many forms. These are some of the types of school violence:

• Physical violence , which includes any kind of physical aggression, the use of weapons, as well as criminal acts like theft or arson.
• Psychological violence , which includes emotional and verbal abuse . This may involve insulting, threatening, ignoring, isolating, rejecting, name-calling, humiliating, ridiculing, rumor-mongering, lying, or punishing another person.
• Sexual violence , which includes sexual harrassment, sexual intimidation, unwanted touching, sexual coercion, and rape .
• Bullying , which can take physical, psychological, or sexual forms and is characterized by repeated and intentional aggression toward another person.
• Cyberbullying , which includes sexual or psychological abuse by people connected through school on social media or other online platforms. This may involve posting false information, hurtful comments, malicious rumors, or embarrassing photos or videos online. Cyberbullying can also take the form of excluding someone from online groups or networks.

Causes of School Violence

There often isn’t a simple, straightforward reason why someone engages in school violence. A child may have been bullied or rejected by a peer, may be under a lot of academic pressure, or may be enacting something they’ve seen at home, in their neighborhood, on television, or in a video game.

These are some of the risk factors that can make a child more likely to commit school violence:

• Poor academic performance
• Prior history of violence
• Hyperactive or impulsive personality
• Mental health conditions
• Witnessing or being a victim of violence
• Alcohol, drug, or tobacco use
• Dysfunctional family dynamic
• Domestic violence or abuse
• Access to weapons
• Delinquent peers
• Poverty or high crime rates in the community

It’s important to note that the presence of these factors doesn’t necessarily mean that the child will engage in violent behavior.

Impact of School Violence

Below, Dr. Daramus explains how school violence can affect children who commit, experience, and witness it, as well as their parents.

Impact on Children Committing Violence

Children who have been victims of violence or exposed to it in some capacity sometimes believe that becoming violent is the only way they‘ll ever be safe.

When they commit violence, they may experience a sense of satisfaction when their emotional need for strength or safety is satisfied. That‘s short-lived however, because they start to fear punishment or retribution, which triggers anger that can sometimes lead to more violence if they’re scared of what might happen to them if they don’t protect themselves. 

Children need help to try and break the cycle; they need to understand that violence can be temporarily satisfying but that it leads to more problems.

Impact on Children Victimized by School Violence

Victims of school violence may get physically injured and experience cuts, scrapes, bruises, broken bones, gunshot wounds, concussions, physical disability, or death.

Emotionally speaking, the child might experience depression , anxiety, or rage. Their academic performance may suffer because it can be hard to focus in school when all you can think about is how to avoid being hurt again.

School violence is traumatic and can cause considerable psychological distress. Traumatic experiences can be difficult for adults too; however, when someone whose brain is not fully developed yet experiences trauma, especially if it’s over a long time, their brain can switch to survival mode, which can affect their attention, concentration, emotional control, and long-term health. 

According to a 2019 study, children who have experienced school violence are at risk for long-term mental and physical health conditions, including attachment disorders, substance abuse, obesity, diabetes, cancer, heart disease, and respiratory conditions.

The more adverse childhood experiences someone has, the greater the risk to their physical and mental health as an adult.

Impact on Children Who Witness School Violence

Children who witness school violence may feel guilty about seeing it and being too afraid to stop it. They may also feel threatened, and their brain may react in a similar way to a child who has faced school violence.

Additionally, when children experience or witness trauma , their basic beliefs about life and other people are often changed. They no longer believe that the world is safe, which can be damaging to their mental health.

For a child to be able to take care of themselves as they get older, they need to first feel safe and cared for. Learning to cope with threats is an advanced lesson that has to be built on a foundation of feeling safe and self-confident.

Children who have experienced or witnessed school violence can benefit from therapy, which can help them process the trauma, regulate their emotions, and learn coping skills to help them heal.

Impact on Parents

Parents react to school violence in all kinds of ways. Some parents encourage their children to bully others, believing that violence is strength. Some try to teach their children how to act in a way that won’t attract bullying or other violence, but that never works and it may teach the child to blame themselves for being bullied. 

Others are proactive and try to work with the school or challenge the school if necessary, to try and keep their child safe. 

It can be helpful to look out for warning signs of violence, which can include:

• Talking about or playing with weapons of any kind
• Harming pets or other animals
• Threatening or bullying others
• Talking about violence, violent movies, or violent games
• Speaking or acting aggressively

It’s important to report these signs to parents, teachers, or school authorities. The child may need help and support, and benefit from intervention .

Preventing School Violence

Dr. Daramus shares some steps that can help prevent school violence:

• Report it to the school: Report any hint of violent behavior to school authorities. Tips can be a huge help in fighting school violence. Many schools allow students to report tips anonymously.
• Inform adults: Children who witness or experience violence should keep telling adults (parents, teachers, and counselors) until someone does something. If an adult hears complaints about a specific child from multiple people, they may be able to protect other students and possibly help the child engaging in violence to learn different ways.
• Reach out to people: Reach out to children or other people at the school who seem to be angry or upset, or appear fascinated with violence. Reach out to any child, whether bullied, bullying, or neither, who seems to have anxiety, depression, or trouble managing emotions. Most of the time the child won’t be violent, but you’ll have helped them anyway by being supportive.

A Word From Verywell

School violence can be traumatic for everyone involved, particularly children. It’s important to take steps to prevent it because children who witness or experience school violence may suffer physical and mental health consequences that can persist well into adulthood.

Centers for Disease Control and Prevention. Preventing school violence .

UNESCO. What you need to know about school violence and bullying .

UNESCO. School violence and bullying .

Nemours Foundation. School violence: what students can do .

Ehiri JE, Hitchcock LI, Ejere HO, Mytton JA. Primary prevention interventions for reducing school violence . Cochrane Database Syst Rev . 2017;2017(3):CD006347. doi:10.1002/14651858.CD006347.pub2

Centers for Disease Control and Prevention. Understanding school violence .

Ferrara P, Franceschini G, Villani A, Corsello G. Physical, psychological and social impact of school violence on children . Italian Journal of Pediatrics . 2019;45(1):76. doi:10.1186/s13052-019-0669-z

By Sanjana Gupta Sanjana is a health writer and editor. Her work spans various health-related topics, including mental health, fitness, nutrition, and wellness.

What you need to know about school violence and bullying

Bullying in schools deprives millions of children and young people of their fundamental right to education. A recent UNESCO report revealed that more than 30% of the world's students have been victims of bullying, with devastating consequences on academic achievement, school dropout, and physical and mental health.

The world is marking the first International Day against Violence and Bullying at School Including Cyberbullying , on 5 November. Here is what you need to know about school violence and bullying.

What is school violence?

School violence refers to all forms of violence that takes place in and around schools and is experienced by students and perpetrated by other students, teachers and other school staff. This includes bullying and cyberbullying. Bullying is one of the most pervasive forms of school violence, affecting 1 in 3 young people.

What forms may school violence take?

Based on existing international surveys that collect data on violence in schools, UNESCO recognizes the following forms of school violence:

• Physical violence, which is any form of physical aggression with intention to hurt perpetrated by peers, teachers or school staff.
• Psychological violence as verbal and emotional abuse, which includes any forms of isolating, rejecting, ignoring, insults, spreading rumors, making up lies, name-calling, ridicule, humiliation and threats, and psychological punishment.
• Sexual violence, which includes intimidation of a sexual nature, sexual harassment, unwanted touching, sexual coercion and rape, and it is perpetrated by a teacher, school staff or a schoolmate or classmate.
• Physical bullying, including hitting, kicking and the destruction of property;
• Psychological bullying, such as teasing, insulting and threatening; or relational, through the spreading of rumours and exclusion from a group; and
• Sexual bullying, such as making fun of a victim with sexual jokes, comments or gestures, which may be defined as sexual ‘harassment’ in some countries.
• Cyberbullying is a form of psychological or sexual bullying that takes place online. Examples of cyberbullying include posting or sending messages, pictures or videos, aimed at harassing, threatening or targeting another person via a variety of media and social media platforms. Cyberbullying may also include spreading rumours, posting false information, hurtful messages, embarrassing comments or photos, or excluding someone from online networks or other communications.

Who perpetrates school violence?

School violence is perpetrated by students, teachers and other school staff. However, available evidence shows that violence perpetrated by peers is the most common.

What are the main reasons why children are bullied?

All children can be bullied, yet evidence shows that children who are perceived to be “different” in any way are more at risk. Key factors include physical appearance, ethnic, linguistic or cultural background, gender, including not conforming to gender norms and stereotypes; social status and disability.

What are the consequences of school violence?

Educational consequences: Being bullied undermines the sense of belonging at school and affects continued engagement in education. Children who are frequently bullied are more likely to feel like an outsider at school, and more likely to want to leave school after finishing secondary education. Children who are bullied have lower academic achievements than those who are not frequently bullied.

Health consequences: Children’s mental health and well-being can be adversely impacted by bullying. Bullying is associated with higher rates of feeling lonely and suicidal, higher rates of smoking, alcohol and cannabis use and lower rates of self-reported life satisfaction and health. School violence can also cause physical injuries and harm.

What are the linkages between school violence and bullying, school-related gender-based violence and violence based on sexual orientation and gender identity or expression?

School violence may be perpetrated as a result of gender norms and stereotypes and enforced by unequal power dynamics and is therefore referred to as school-related gender-based violence. It includes, in particular, a specific type of gender-based violence that is linked to the actual or perceived sexual orientation and gender identity or expression of victims, including homophobic and transphobic bullying. School-related gender-based violence is a significant part of school violence that requires specific efforts to address.

Does school-related gender-based violence refer to sexual violence against girls only?         

No. School-related gender-based violence refers to all forms of school violence that is based on or driven by gender norms and stereotypes, which also includes violence against and between boys.

Is school violence always gender-based?           

There are many factors that drive school violence. Gender is one of the significant drivers of violence but not all school violence is based on gender. Moreover, international surveys do not systematically collect data on the gendered nature of school violence, nor on violence based on sexual orientation and gender identity or expression. 

Based on the analysis of global data, there are no major differences in the prevalence of bullying for boys and girls. However, there are some differences between boys and girls in terms of the types of bullying they experience. Boys are much more exposed to physical bullying, and to physical violence in general, than girls. Girls are slightly more exposed to psychological bullying, particularly through cyberbullying. According to the same data, sexual bullying the same proportion of boys and girls. Data coming from different countries, however, shows that girls are increasingly exposed to sexual bullying online.

How does UNESCO help prevent and address school violence and bullying?

The best available evidence shows that responses to school violence and bullying that are effective should be comprehensive and include a combination of policies and interventions. Often this comprehensive response to school violence and bullying is referred to as a whole-school approach. Based on an extensive review of existing conceptual frameworks that describe that whole-school approach, UNESCO has identified nine key components of a response that goes beyond schools and could be better described as a whole-education system or whole-education approach.  These components are the following:

• Strong political leadership and robust legal and policy framework to address school violence and bullying;
• Training and support for teachers on school violence and bullying prevention and positive classroom management
• Curriculum, learning & teaching to promote, a caring (i.e. anti- school violence and bullying) school climate and students’ social and emotional skills
• A safe psychological and physical school and classroom environment
• Reporting mechanisms for students affected by school violence and bullying, together with support and referral services
• Involvement of all stakeholders in the school community including parents
• Student empowerment and participation
• Collaboration and partnerships between the education sector and a wide range of partners (other government sectors, NGOs, academia)
• Evidence: monitoring of school violence and bullying and evaluation of responses

More on UNESCO’s work to prevent and address school violence and bullying

Read UNESCO's publication Behind the numbers: Ending school violence and bullying

Photo: Eakachai Leesin/Shutterstock.com

More on this subject

Other recent articles

Violence in Schools Is a National Crisis

Aggression is often triggered by common directives and disciplinary practices..

Posted February 20, 2020 | Reviewed by Lybi Ma

School violence is a problem both on and off school grounds, with reports of more than 20 episodes of gun violence at or near U.S. school sporting events since mid-August. In January of this year, three school shootings in California and Texas resulted in four people killed or injured, including a 9-year-old girl on the school playground.

The most recent national report on Indicators of School Crime and Safety reveals that of about 3.8 million teachers, 10 percent (or 373,900) teachers reported that students had threatened them with injury. Another 6 percent, or 220,300 teachers, reported that a student had physically attacked them.

School violence represents a national crisis. Schools struggle to address this problem, and many educators lack the training and support to effectively prevent and address violence. Our research is diving into violence against teachers and strategies to solve the issue.

As a 25-year scholar on school violence and a member of the American Psychological Association National Task Force on Violence Against Teachers , I, along with colleagues, anonymously surveyed over 3,000 pre-kindergarten through 12th-grade teachers across the United States to better understand the violence that teachers experience in schools.

We found that while verbal aggression is the most common form, 44 percent of teachers reported experiencing at least one physical offense . Physical offenses are often serious, and 9 percent of teachers reported a doctor visit following the injury. Many teachers are physically victimized by students, parents, and colleagues , making the workplace a toxic environment.

In our study, one 63-year-old female teacher in an urban public elementary school in Maryland reported: "I was attacked from behind—a large object was thrown that hit me in the upper back and caused whiplash injury to my neck. This is the fourth time in four years that I have been hurt badly enough to need to see a doctor."

A 63-year-old male teacher in an urban middle school in Iowa reported: "Had student pull a gun on me in the parking lot and threaten to shoot me... The student got angry; he was caught cheating."

In another instance, a 57-year-old female, suburban public middle school teacher from Louisiana indicated: "I have logged in knives, drugs, brass knuckles, and the administration and deputies confiscated the AK-47… One student had detailed out, on every page, how he intended to kill me and bring a gun to school."

Studies that go beyond the prevalence of violence directed at teachers are limited, and researchers know little about the triggers and consequences surrounding these incidents.

The Task Force recently published a study on physical aggression toward teachers using an antecedent‐behavior‐consequence framework. Among 193 U.S. teachers, the common antecedents included breaking up fights, discipline, and directives. Common consequences included student removal, police or legal involvement, school staff involvement, positive outcomes, and inaction.

Given that physical aggression is often triggered by common teacher directives and school disciplinary practices, educators are frustrated. Schools are engaging in practices with the hope of improving safety and need research to examine their effectiveness.

I am chairing a newly created APA National Task Force on Violence Against Educators that will build upon the previous APA Task Force findings and broaden our study to include multiple school stakeholders. We are partnering with many national educational organizations to assess school violence against educators, para-professionals, administrators, school psychologists, and school social workers, as well as the effectiveness of various policies, practices, and training to inform the next generation of research on this important issue.

Given growing concerns about school violence, student protesters are calling for action in Iowa. Ohio created a new School Safety Center. Johns Hopkins University in Maryland established a new Center for Safe and Healthy Schools . Virginia proposed legislation to improve teacher training through adding instructional requirements on positive behavioral interventions, crisis prevention and de-escalation, and the proper use of and prevention of physical restraints.

Reducing school violence requires more effective action, research, training, and policy to improve safety, school climate, and well-being for students and educators.

McMahon, S.D., Martinez, A., Espelage, D., Rose, C., Reddy, L.A., Lane, K., Anderman, E. M., Reynolds, C. R., Jones, A., & Brown, V. (2014). Violence directed against teachers: Results from a national survey. Psychology in the Schools, 51 , 753-766. DOI: 10.1002/pits.21777

McMahon, S.D., Peist, E., Davis, J.O., & Bare, K., Martinez, A., Reddy, L.A., Espelage, D.L., & Anderman, E.M. (2020). Physical aggression toward teachers: Antecedents, behaviors, and consequences. Aggressive Behavior, 46 , 116-126. https://doi.org/10.1002/ab.21870

Susan Dvorak McMahon, Ph.D., is a Vincent DePaul Professor of Clinical and Community Psychology and Associate Dean for Research at DePaul University.

• Find a Therapist
• Find a Treatment Center
• Find a Psychiatrist
• Find a Support Group
• Find Teletherapy
• United States
• Brooklyn, NY
• Chicago, IL
• Houston, TX
• Los Angeles, CA
• New York, NY
• Portland, OR
• San Diego, CA
• San Francisco, CA
• Seattle, WA
• Washington, DC
• Asperger's
• Bipolar Disorder
• Chronic Pain
• Eating Disorders
• Passive Aggression
• Personality
• Goal Setting
• Positive Psychology
• Stopping Smoking
• Low Sexual Desire
• Relationships
• Child Development
• Therapy Center NEW
• Diagnosis Dictionary
• Types of Therapy

Understanding what emotional intelligence looks like and the steps needed to improve it could light a path to a more emotionally adept world.

• Coronavirus Disease 2019
• Affective Forecasting
• Neuroscience

Fast Fact: Preventing School Violence

School violence is violence that occurs in the school setting. It describes violent acts that disrupt learning and have a negative effect on students, schools, and the broader community. School is the location where the violence occurs, not a type of violence.

Examples of school violence include:

• Bullying and cyberbullying
• Fighting (e.g., punching, slapping, kicking)
• Gang violence
• Sexual violence

Places school violence occurs:

• On school property
• On the way to or from school
• During a school-sponsored event
• On the way to or from a school-sponsored event

In 2019, CDC’s nationwide Youth Risk Behavior Survey (YRBS) was administered to high school students across the United States. According to YRBS results  from 13, 677 students:

• About 1 in 5 high school students reported being bullied on school property in the last year.
• 8% of high school students had been in a physical fight on school property one or more times during the 12 months before the survey.
• More than 7% of high school students had been threatened or injured with a weapon (for example, a gun, knife, or club) on school property one or more times during the 12 months before the survey.
• Almost 9% of high school students had not gone to school at least 1 day during the 30 days before the survey because they felt they would be unsafe at school or on their way to or from school.

All students have the right to learn in a safe school environment. The good news is school violence can be prevented. Many factors contribute to school violence. Preventing school violence requires addressing the factors that put people at risk for or protect them from violence. Research shows that prevention efforts by teachers, administrators, parents, community members, and even students can reduce violence and improve the school environment.

CDC developed Resources for Action , formerly known as, “technical packages,” to help communities and states prioritize prevention strategies based on the best available evidence. The strategies and approaches in the Resources for Action are intended to shape individual behaviors as well as the relationship, family, school, community, and societal factors that influence risk and protective factors for violence. They are meant to work together and to be used in combination in a multi-level, multi-sector effort to prevent violence.

See  Youth Violence Resources  for articles, publications, data sources, and prevention resources for school violence.

To receive email updates about this page, enter your email address:

• Data & Statistics (WISQARS)
• Opioid Overdose Prevention
• Traumatic Brain Injury
• Motor Vehicle Safety

Exit Notification / Disclaimer Policy

• The Centers for Disease Control and Prevention (CDC) cannot attest to the accuracy of a non-federal website.
• Linking to a non-federal website does not constitute an endorsement by CDC or any of its employees of the sponsors or the information and products presented on the website.
• You will be subject to the destination website's privacy policy when you follow the link.
• CDC is not responsible for Section 508 compliance (accessibility) on other federal or private website.

• High contrast
• Press Centre

Search UNICEF

Protecting children from violence in school, every child has the right to go to school and learn, free from fear..

• Violence against children
• Violence in school

Every child has the right to go to school free from fear. When schools provide quality, inclusive and safe education, children can learn, build friendships and gain the critical skills they need to navigate social situations. In the best circumstances, school puts children on the path to a promising future.

But for too many girls and boys worldwide, school is where they experience violence. Bullying, harassment, verbal abuse, sexual abuse and exploitation, corporal punishment and other forms of humiliation can come at the hands of a peer, a teacher or even a school authority. Many children also experience school violence associated with gang culture, weapons and fighting.

Far from a haven for learning and community, school can be a place of bullying, sexual harassment, corporal punishment, verbal abuse and other forms of violence.

Violence in schools can have serious effects on children’s psychological and physical health.

Children who are subjected to violence may experience physical injury, sexually transmitted infections, depression, anxiety, post-traumatic stress disorder (PTSD) and suicidal thoughts. They may also begin to exhibit risky, aggressive and anti-social behaviour. Children who grow up around violence have a greater chance of replicating it for a new generation of victims.

At its most extreme, violence in and around schools can be deadly. For the tens of millions of children and adolescents living in conflict-affected areas, school too often becomes the front line.

What’s more, violence in school can reduce school attendance, lower academic performance and increase drop-out rates. This has devastating consequences for the success and prosperity of children, their families and entire communities.

• Globally, half of students aged 13–15 – some 150 million – report experiencing peer-to-peer violence in and around school.
• Slightly more than 1 in 3 students between the ages of 13 and 15 experience bullying, and about the same proportion are involved in physical fights.
• Around 720 million school-aged children live in countries where they are not fully protected by law from corporal punishment at school.
• Between 2005 and 2020, the United Nations verified more than 13,900 incidents of attacks, including direct attacks or attacks where there has not been adequate distinction between civilian and military objectives, on educational and medical facilities and protected persons, including pupils and hospitalised children, and health and school personnel. 

UNICEF’s response

UNICEF works with governments, schools, teachers, families, children and young people to prevent and respond to violence in schools. We help governments and partners:

• Adopt laws prohibiting corporal punishment and other forms of violence.
• Develop codes of conduct and other safeguarding measures in schools.
• Set up confidential and safe reporting mechanisms in schools.
• Establish a referral mechanism for response services, and monitor and collect data on violence in schools.
• Train teachers and school staff on positive discipline, classroom management and peaceful conflict resolution.
• Develop and implement life skills and social and emotional learning programmes to build the resilience and protective capacity of children and youth.
• Research, monitor and collect data on violence in schools.

As part of Safe to Learn – an inter-agency and multi-country initiative dedicated to ending violence in and around schools – UNICEF also works to increase the protection of children, improve learning outcomes, better leverage investments in education, and raise awareness of violence in schools.

More from UNICEF

DR Congo: Children killed, injured, abducted, and face sexual violence in conflict at record levels for third consecutive year – UNICEF

On World Poetry Day, hundreds of children affected by conflict and war share Poems for Peace

UNICEF Goodwill Ambassador Angélique Kidjo celebrates resilience of girls and young people during visit to Benin

2 in 3 children in Latin America and the Caribbean experience violence at home

UNICEF Humanitarian Practice: COVID-19 Technical Guidance

Action to end violence against children in schools: review of programme interventions illustrating actions to address violence against children in and around schools, social and behaviour change to address violence against children: technical guidance, social and behaviour change strategies for addressing violence against children in and around schools: case studies and lessons learned, safe to learn: global programmatic framework & benchmarking tool: from call to action to programme responses, safe to learn: safe to learn diagnostic exercises in nepal, pakistan, south sudan and uganda synthesis report  , safe to learn: diagnostic tool, safe to learn:  safe to learn  in action how nepal, pakistan, south sudan and uganda are meeting the challenge of ending violence in schools, school-based violence prevention: a practical handbook, child-friendly schools manual, an everyday lesson: #endviolence in schools, behind the numbers: ending school violence and bullying, global guidance on addressing school-related gender-based violence, tackling violence in schools: bridging the gap between standards and practice, ending the torment: tackling bullying from the schoolyard to cyberspace, a rigorous review of global research evidence on policy and practice on school-related gender-based violence, preventing bullying: the role of public health and safety professionals, the campaign to stop violence in schools: third progress report, protecting children from bullying: report of the secretary-general, violence against children in education settings in south asia, violence against children: united nations secretary-general’s study, save the children global report 2017: ending violence in childhood.

Last updated 27 August 2021

Why It's a Bad Idea to Tell Students Words Are Violence

A claim increasingly heard on campus will make them more anxious and more willing to justify physical harm.

Of all the ideas percolating on college campuses these days, the most dangerous one might be that speech is sometimes violence. We’re not talking about verbal threats of violence, which are used to coerce and intimidate, and which are illegal and not protected by the First Amendment. We’re talking about speech that is deemed by members of an identity group to be critical of the group, or speech that is otherwise upsetting to members of the group. This is the kind of speech that many students today refer to as a form of violence. If Milo Yiannopoulos speaks on the University of California, Berkeley, campus, is that an act of violence?

Recently, the psychologist Lisa Feldman Barrett, a highly respected emotion researcher at Northeastern University, published an essay in The New York Times titled, “ When is speech violence ?” She offered support from neuroscience and health-psychology research for students who want to use the word “violence” in this expansive way. The essay made two points that we think are valid and important, but it drew two inferences from those points that we think are invalid.

First valid point: Chronic stress can cause physical damage. Feldman Barrett cited research on the ways that chronic (not short-term) stressors “can make you sick , alter your brain —even kill neurons —and shorten your life .” The research here is indeed clear.

Recommended Reading

The Coddling of the American Mind

The Quiet Cruelty of When Harry Met Sally

The Hidden Costs of Living Alone

First invalid inference: Feldman Barrett used these empirical findings to advance a syllogism: “If words can cause stress, and if prolonged stress can cause physical harm, then it seems that speech—at least certain types of speech—can be a form of violence.” It is logically true that if A can cause B and B can cause C, then A can cause C. But following this logic, the resulting inference should be merely that words can cause physical harm, not that words are violence. If you’re not convinced, just re-run the syllogism starting with “gossiping about a rival,” for example, or “giving one’s students a lot of homework.” Both practices can cause prolonged stress to others, but that doesn’t turn them into forms of violence.

Feldman Barrett’s second valid point lies in her argument that young people are antifragile —they grow from facing and overcoming adversity:

Offensiveness is not bad for your body and brain. Your nervous system evolved to withstand periodic bouts of stress, such as fleeing from a tiger, taking a punch or encountering an odious idea in a university lecture. Entertaining someone else’s distasteful perspective can be educational. ... When you’re forced to engage a position you strongly disagree with, you learn something about the other perspective as well as your own. The process feels unpleasant, but it’s a good kind of stress — temporary and not harmful to your body — and you reap the longer-term benefits of learning.

Feldman Barrett could have gone a step further: This “good kind of stress” isn’t just “not harmful,” it also sometimes makes an individual stronger and more resilient . The next time that person faces a similar situation, she’ll experience a milder stress response because it is no longer novel, and because her coping repertoire has grown. This was the argument at the heart of our 2015 essay in The Atlantic , “The Coddling of the American Mind .” We worried that colleges were making students more fragile—more easily harmed—by trying to protect them from the sorts of small and brief offensive experiences that Feldman Barrett is talking about.

Feldman Barrett then contrasted brief experiences of offensiveness with chronic stressors:

What’s bad for your nervous system, in contrast, are long stretches of simmering stress. If you spend a lot of time in a harsh environment worrying about your safety, that’s the kind of stress that brings on illness and remodels your brain. That’s also true of a political climate in which groups of people endlessly hurl hateful words at one another, and of rampant bullying in school or on social media. A culture of constant, casual brutality is toxic to the body, and we suffer for it.

We agree. But what, then, are the implications for college campuses?

In Feldman Barrett’s second invalid inference, she writes:

That’s why it’s reasonable, scientifically speaking, not to allow a provocateur and hatemonger like Milo Yiannopoulos to speak at your school. He is part of something noxious, a campaign of abuse. There is nothing to be gained from debating him, for debate is not what he is offering.

But wait, wasn’t Feldman Barrett’s key point the contrast between short- and long-term stressors? What would have happened had Yiannopoulos been allowed to speak at Berkeley? He would have faced a gigantic crowd of peaceful protesters, inside and outside the venue. The event would have been over in two hours. Any students who thought his words would cause them trauma could have avoided the talk and left the protesting to others. Anyone who joined the protests would have left with a strong sense of campus solidarity. And most importantly, all Berkeley students would have learned an essential lesson for life in 2017: How to encounter a troll without losing one’s cool. (The goal of a troll, after all, is to make people lose their cool .)

Feldman Barrett’s argument only makes sense if Yiannopoulos’s speech is interpreted as one brief episode in a long stretch of “simmering stress” on campus. The argument works only if Berkeley students experience their school as a “harsh environment,” a “culture of constant, casual brutality” in which they are chronically “worrying about [their] safety.” Maybe that is the perception of some students. But if so, is the solution to change the school or to change the perception?

Aggressive and even violent protests have erupted at some of the country’s most progressive schools, such as Berkeley, Middlebury College , and Evergreen State College . Are these schools brutal and toxic environments for members of various identity groups? Or has a set of new ideas on campus taught students to see oppression and violence wherever they look? If students are repeatedly told that numerical disparities are proof of systemic discrimination , and a clumsy or insensitive question is an act of aggression (a “microaggression”) , and words are sometimes acts of violence that will shorten your life , then it begins to make sense that they would worry about their safety, chronically, even within some of America’s most welcoming and protective institutions.

We are not denying that college students encounter racism and other forms of discrimination on campus, from individuals or from institutional systems. We are, rather, pointing out a fact that is crucial in any discussion of stress and its effects: People do not react to the world as it is; they react to the world as they interpret it, and those interpretations are major determinants of success and failure in life. As we said in our Atlantic article :

Rather than trying to protect students from words and ideas that they will inevitably encounter, colleges should do all they can to equip students to thrive in a world full of words and ideas that they cannot control. One of the great truths taught by Buddhism (and Stoicism, Hinduism, and many other traditions) is that you can never achieve happiness by making the world conform to your desires. But you can master your desires and habits of thought. This, of course, is the goal of cognitive behavioral therapy.

We wrote those words in early 2015. We were responding to stories from across the country about new demands that students were making for protection from the kinds of offensiveness that Feldman Barrett says are “not bad for your body or brain.” We explained why we thought that widespread adoption of trigger warnings, safe spaces, and microaggression training would backfire. Rather than keeping students safe from harm, a culture of “safety” teaches students to engage in some of the same cognitive distortions that cognitive-behavioral therapy tries to eliminate. Distortions such as “emotional reasoning,” “catastrophizing,” and “dichotomous thinking,” we noted, are associated with anxiety, depression, and difficulty coping. We think our argument is much stronger today, for two reasons.

First, our article was published in August of 2015, a few months before a wave of campus protests began at Missouri, Yale, and dozens of other schools. Those protesters usually demanded that their universities implement an array of policies designed to keep students “safer” from offense—policies such as microaggression training supplemented by the creation of systems for reporting and punishing microaggressors , along with the creation of more ethnic- or identity-based centers. We expect that these policies—whose effectiveness is not supported by empirical evidence —will, in the long run, lead students to feel even less “safe” on campus than they did in 2015, because they may increase the number of offenses perceived while heightening feelings of identity-based division and victimization . Some evidence also suggests that diversity training, when not carefully and sensitively implemented, can create a backlash , which amplifies tensions.

Second, we wrote our article at a time that saw hints of a mental-health crisis on campuses, but no conclusive survey evidence. Two years later, the evidence is overwhelming. The social psychologist Jean Twenge has just written a book, titled iGen (which is short for “internet generation”), in which she analyzes four large national datasets that track the mental health of teenagers and college students. When the book is released in August, Americans will likely be stunned by her findings. Graph after graph shows the same pattern: Lines drift mildly up or down across the decades as baby boomers are followed by Gen-X, which is followed by the millennials. But as soon as the data includes iGen—those born after roughly 1994—the rates of anxiety, depression, loneliness, and suicide spike upward.

Is iGen so different from the millennials because the former faces more chronic, long-term stress? Have the country’s colleges suddenly become brutal, toxic places, increasingly hostile to members of various identity groups? Some would argue, as Twenge does, that social media changed the nature of iGen’s social interactions. But if social media is the biggest cause of the mental-health crisis then the solution lies in changing the nature or availability of social media for teenagers. Making the offline world “safer” by banning the occasional stress-inducing speaker will not help.

We think the mental-health crisis on campus is better understood as a crisis of resilience. Since 2012, when members of iGen first began entering college, growing numbers of college students have become less able to cope with the challenges of campus life , including offensive ideas, insensitive professors, and rude or even racist and sexist peers. Previous generations of college students learned to live with such challenges in preparation for success in the far more offense-filled world beyond the college gates. As Van Jones put it in response to a question by David Axelrod about how progressive students should react to ideologically offensive speakers on campus:

I don’t want you to be safe, ideologically. I don’t want you to be safe, emotionally. I want you to be strong. That’s different. I’m not going to pave the jungle for you. Put on some boots, and learn how to deal with adversity. I’m not going to take all the weights out of the gym; that’s the whole point of the gym. This is the gym.

This is why the idea that speech is violence is so dangerous. It tells the members of a generation already beset by anxiety and depression that the world is a far more violent and threatening place than it really is. It tells them that words, ideas, and speakers can literally kill them. Even worse: At a time of rapidly rising political polarization in America, it helps a small subset of that generation justify political violence. A few days after the riot that shut down Yiannopoulos’s talk at Berkeley, in which many people were punched, beaten, and pepper sprayed by masked protesters, the main campus newspaper ran five op-ed essays by students and recent alumni under the series title “ Violence as self defense .” One excerpt: “Asking people to maintain peaceful dialogue with those who legitimately do not think their lives matter is a violent act.”

The implication of this expansive use of the word “violence” is that “we” are justified in punching and pepper-spraying “them,” even if all they did was say words. We’re just defending ourselves against their “violence.” But if this way of thinking leads to actual violence, and if that violence triggers counter-violence from the other side (as happened a few weeks later at Berkeley ), then where does it end? In the country’s polarized democracy, telling young people that “words are violence” may in fact lead to a rise in real, physical violence.

Free speech, properly understood, is not violence. It is a cure for violence.

In his 1993 book Kindly Inquisitors , the author Jonathan Rauch explains that freedom of speech is part of a system he calls “Liberal Science”—an intellectual system that arose with the Enlightenment and made the movement so successful. The rules of Liberal Science include: No argument is ever truly over, anyone can participate in the debate, and no one gets to claim special authority to end a question once and for all. Central to this idea is the role of evidence, debate, discussion, and persuasion. Rauch contrasts Liberal Science with the system that dominated before it—the “Fundamentalist” system—in which kings, priests, oligarchs, and others with power decide what is true, and then get to enforce orthodoxy using violence.

Liberal Science led to the radical social invention of a strong distinction between words and actions, and though some on campus question that distinction today, it has been one of the most valuable inventions in the service of peace, progress, and innovation that human civilization ever came up with. Freedom of speech is the eternally radical idea that individuals will try to settle their differences through debate and discussion, through evidence and attempts at persuasion, rather than through the coercive power of administrative authorities—or violence.

To be clear, when we refer to “free speech,” we are not talking about things like threats, intimidation, and incitement. The First Amendment provides categorical exceptions for those because such words are linked to actual physical violence. The First Amendment also excludes harassment—when words are used in a directed pattern of discriminatory behavior.

But the extraordinary body of legal reasoning that has developed around the First Amendment also recognizes that universities are different from other settings. In a 2010 decision by the U.S. Court of Appeals for the Ninth Circuit— Rodriguez v. Maricopa County Community College District —Chief Judge Alex Kozinski noted “...the urge to censor is greatest where debate is most disquieting and orthodoxy most entrenched…” He then explained the special nature of universities, using terms that illustrate Rauch’s Liberal Science:

The right to provoke, offend, and shock lies at the core of the First Amendment. This is particularly so on college campuses. Intellectual advancement has traditionally progressed through discord and dissent, as a diversity of views ensures that ideas survive because they are correct, not because they are popular. Colleges and universities—sheltered from the currents of popular opinion by tradition, geography, tenure and monetary endowments—have historically fostered that exchange. But that role in our society will not survive if certain points of view may be declared beyond the pale.

In sum, it was a radical enlightenment idea to tolerate the existence of dissenters, and an even more radical idea to actually engage with them. Universities are—or should be—the preeminent centers of Liberal Science. They have a duty to foster an intellectual climate that separates true ideas from popular but fallacious ones.

The conflation of words with violence is not a new or progressive idea invented on college campuses in the last two years. It is an ancient and regressive idea. Americans should all be troubled that it is becoming popular again—especially on college campuses, where it least belongs.

U.S. Supreme Court Has Schools in Mind as It Weighs What ‘True Threats’ Are

• Share article

The U.S. Supreme Court on Wednesday weighed when statements constitute “true threats” that are not protected by the First Amendment, and the justices had potentially menacing speech involving schools on their minds.

“Let’s imagine this example,” Justice Amy Coney Barrett said during arguments in Counterman v. Colorado . “A teenager in a high school says something like, you know, ‘I’m going to shoot this place down,’ and it’s devoid of all context.”

The school, taking the threat seriously, Barrett said, “wants the kid to be barred from the grounds or wants him to be suspended for a few days so they can assess the threat. … Could the school do that just based on that one statement?”

John P. Elwood, a Washington lawyer representing a man sentenced to four-and-a-half years in prison for sending disturbing Facebook messages that the sender contends were not meant to be threatening, said he believed a school could discipline a student in that situation.

“Schools have extra leeway, and schools are a whole ball of wax” different from law enforcement treatment of such speech, Elwood said.

Left unsaid in that response is that law enforcement often quickly gets involved in school threats, and students frequently face criminal prosecution in addition to school discipline. School administrators and the legal community have been seeking better guidance on when speech-based threats may be punished either under school discipline or criminal or juvenile courts. The outcome of the Colorado case may provide some guidance on those questions.

Justice Brett M. Kavanaugh asked a U.S. Department of Justice lawyer whether there were any statistics or studies about school shootings or other incidents of violence “that perhaps could have been prevented if threats had been taken more seriously beforehand?”

Eric J. Feigin, the deputy U.S. solicitor general arguing in support of Colorado’s prosecution of the Facebook threat-maker, said he didn’t have any numbers to offer, but the question reflected the experience “that there is frequently after one of these horrific incidents some question of … ‘you know, why didn’t you intervene, why didn’t you respond earlier?’”

“It is very important that the [government] have some ability to intervene at an earlier stage,” Feigin said. “And legislatures shouldn’t be precluded from making the judgment that those kinds of harms are more important, particularly in the case of reckless defendants who decide that they will inspire fear in others to further their own selfish ends.”

The importance of the “reasonable person”

The case before the justices does not involve a school threat but postings on Facebook by Billy Raymond Counterman, who became enthralled with a singer-songwriter identified in court papers as C.W. Counterman sent her hundreds of messages and sometimes feigned friendship or intimacy that simply did not exist, and at other times sent messages that she perceived as menacing.

Counterman was charged and convicted under a state law against stalking. Counterman’s lawyers say he suffers from mental illness and never intended any threats. He was barred from submitting any evidence that he believed C.W. was corresponding with him. The prosecution and a trial court applied an objective standard requiring the jury to convict if it found that Counterman’s messages “would cause a reasonable person to suffer serious emotional distress.”

The question before the Supreme Court is whether it is enough to show only that an objective “reasonable person” would regard the statement in question as a threat of violence, to which Colorado contends, or whether the government must show that the speaker subjectively knew or intended the threatening nature of the statement, as Counterman’s lawyer argued.

“Criminalizing misunderstanding is especially dangerous in an age when so much communication occurs on social media, which brings together strangers in an environment that removes much of the context that gives words meaning,” Elwood told the justices. “And it chills expression by imposing prison time on speakers who do not tailor their views to suit their audience.”

Colorado Attorney General Philip J. Weiser, arguing to uphold the conviction, said that an “objective, context-driven inquiry means that this test won’t criminalize a joke taken the wrong way, political advocacy, or hyperbole. It thus protects statements that contribute to the marketplace of ideas.”

In Colorado’s merits brief , Weiser noted that threats on the 20th anniversary of the 1999 mass shooting at Columbine High School in Littleton, Colo., led to hundreds of school closures across Colorado.

“The First Amendment interests of those who are threatened, not just the asserted First Amendment interests of those who make threats, are at stake here,” Weiser says in the brief.

Concerns about “eggshell” sensibilities

The court wrestled with these issues about eight years ago when it considered the case of a Pennsylvania man who made threats on Facebook that included rap lyric-style musings about shooting up an elementary school.

The justices ruled 8-1 in Elonis v. United States in 2015 to toss the federal conviction of Anthony Elonis, but the majority stopped short of making any broad First Amendment rulings about threats on the internet.

Four justices’ seats have turned over since that decision. The overall tone of Wednesday’s arguments showed some skepticism toward Colorado’s case.

The arguments showed some hints that changes outside the court may be affecting how the justices view the issue. There was discussion about whether the “reasonable observer” of the objective standard might be too open to perceiving speech as threatening in the conflict-filled society of 2023.

“Who is the reasonable person?” Barrett asked, wondering whether if it were speech on a college campus, “is it the reasonable college student?”

“Let’s imagine a professor who wants people to understand just how vicious it was to be in a Jim Crow South and puts up behind them on a screen a picture of a burning cross and reads aloud some threats of lynching that were made at the time,” Barrett continued. “Purely educational purpose in the teacher’s mind, but students feel physically threatened, they fear for their safety because they don’t understand it.”

She went on to suggest that a Black student sitting in that classroom might perceive the lesson as more threatening than a white peer.

“We might have differences about who we think are the eggshell audience or not,” Barrett said, in an apparent reference to people with overly delicate sensibilities.

Justice Clarence Thomas addressed the same concern about the objective “reasonable person.”

“We’re more hypersensitive about different things now, and people could feel threatened in different ways,” he said.

And Justice Neil M. Gorsuch referred to professors who issue “trigger warnings” to their students about “difficult” educational content.

“We live in a world in which people are sensitive, and maybe increasingly sensitive,” he said.“Aren’t a lot of things harmful that we talk about—and have to talk about—difficult, offensive to reasonable people? Some of our history could count as that. Some of the court’s cases might even count as that.”

A decision in the case is expected by late June.

Sign Up for EdWeek Update

Edweek top school jobs.

Sign Up & Sign In

When Is a Public School Student’s Online Speech About School Violence Cause for Concern?

I’ve written, here at Justia’s Verdict , a number of times about cases regarding the First Amendment rights of public school students. I think that these cases are especially important because they can be formative, in either positive or negative ways, for the student or students who are directly affected by them, and for their classmates, as they watch the case unfurl.

The leading Supreme Court precedent in the school speech area is Tinker v. Des Moines Indep. Cmty. Sch. Dist. Almost always, in studying the cases governed by Tinker, I’ve found that school administrations tend to overreact regarding controversial but also First-Amendment-protected student speech, and thus, they do not follow Tinker ’s dictates as faithfully as they should.  In such cases, students are therefore often wronged– at least until their parents, or a public interest group such as the ACLU, finds them a good attorney. But a recent case that was heard by the U.S. Court of Appeals for the Ninth Circuit, Wynar v. Douglas County Sch. Dist. , is significantly different from these other school-speech precedents that I’ve previously commented on, as I will explain.

The Facts of Wynar

The Wynar case arose after Nevada public high school sophomore Landon Wynar, while at home, sent a series of Instant Messages (IMs) to his friends on MySpace bragging about the weapons he owned, which included various rifles, including a Russian semi-automatic and a .22 caliber rifle; threatening to shoot specific classmates; and invoking the Virginia Tech Massacre.

Some of those friends, unsettled by Landon’s remarks, rightly alerted school authorities, and as a result, Landon was expelled from school.  His defense, in the expulsion proceeding, was that he was only joking. Although he could have called witnesses in his defense in that proceeding, he did not do so.

The Ninth Circuit panel affirmed the district court’s grant of summary judgment in favor of the school district, despite Landon’s First Amendment defenses. The result here is unsurprising, as Landon’s MySpace messages eventually centered on a future school shooting centering around the dates of Hitler’s birth, the Columbine massacre, and the Virginia Tech Massacre.

Landon challenged the expulsion on First Amendment grounds.  However, the U.S. District Court granted summary judgment in the school’s favor, while also noting that the U.S. Supreme Court has not yet ruled on the law applying to the off-campus speech of public school students when that speech pertains directly to the public school.

However, as the District Court noted, there is precedent in the Ninth Circuit that is closely relevant to the Wynar case.

The Key Ninth Circuit Precedent

In that closely relevant case, LaVine v. Blaine Sch. Dist. , the Ninth Circuit upheld the temporary, emergency expulsion of a student who had written a first-person poem, at home, about a school shooting and a subsequent suicide.  Applying Tinker ’s test, the court found that the poem fulfilled the criteria of Tinker : It constituted speech that might reasonably lead school authorities to forecast substantial disruption of, or material interference with, school activities, or that collides with the rights of other students to be secure and to be let alone.

Does Off-Campus Student Speech Come Within the Tinker Test?

An interesting facet of public school cases, in the age of the Internet, is that students may carry on the same conversation with their peers as they travel from school to home, and vice-versa, via a variety of media.  That reality raises an interesting question as to whether Tinker ’s reach is long enough to reach such conversations.

As the Wynar court notes, a number of federal Circuits (the Second, Fourth, and Eighth) say yes, while the Third and Fifth have left the question open thus far.  My sense is that, in the end, the test will have to relate to the subject matter of the conversation (that is, whether it is school-related or non-school-related), rather than the physical or virtual location of that conversation.

Returning to Wynar itself, there is little doubt that if the messages that Wynar had sent to his friends come under the Tinker test, even despite the fact that they were posted on MySpace, because of their school-related subject matter, then those messages should be deemed to be speech that might reasonably lead school authorities to forecast the substantial disruption of, or the material interference with, school activities, and/or speech that collides with the rights of other students to be secure and to be let alone.

In the end, Landon Wynar lost his case due to a number of bad facts.  Important among them were (1) the fact that, in the court’s eyes, Landon’s MySpace messages could be seen as a plan to attack the school; (2) the fact that Landon possessed weapons that he could use to carry out that possible plan; (3) and the fact that the consequences of such a plan, if they ensued, would be catastrophic.

These three key facts, along with other facts from the case, convinced the Ninth Circuit panel to rule in the school’s favor and thus to uphold the expulsion – and rightly so. Many school speech cases are just that: A dispute about a student’s comments or a student body’s protests, centering on events at the school. Landon’s case, unlike other school speech cases, has much higher stakes, for it has a marked undertone of possible future school violence. Landon needs some counseling now, while he is still young, in order to ensure that he doesn’t someday translate his First Amendment-protected thoughts and comments, over time, into something much more dangerous.

Posted in: Speech and Religion

One response to “When Is a Public School Student’s Online Speech About School Violence Cause for Concern?”

“as they watch the case unspurl.” Is that really something our children should be doing? Furling and unfurling, now, I could see. But watching things unspurl? Maybe best done in private.

by Julie Hilden

• Advisory Board
• Project on Law & Applied Neuroscience
• Juvenile Justice
• Criminal Responsibility
• Tracking New Cases
• Opportunities
• Make a Donation
• Stay Connected

The Center for Law, Brain & Behavior puts the most accurate and actionable neuroscience in the hands of judges, lawyers, policymakers and journalists—people who shape the standards and practices of our legal system and affect its impact on people’s lives. We work to make the legal system more effective and more just for all those affected by the law.

When Is Speech Violence?

By Lisa Feldman Barrett | The New York Times | July 14, 2017

Imagine that a bully threatens to punch you in the face. A week later, he walks up to you and breaks your nose with his fist. Which is more harmful: the punch or the threat?

The answer might seem obvious: Physical violence is physically damaging; verbal statements aren’t. “Sticks and stones can break my bones, but words will never hurt me.”

But scientifically speaking, it’s not that simple. Words can have a  powerful effect on your nervous system . Certain types of adversity, even those involving no physical contact, can  make you sick ,  alter your brain  — even  kill neurons  — and  shorten your life .

Your body’s immune system includes little proteins called proinflammatory cytokines that cause inflammation when you’re physically injured. Under certain conditions, however, these cytokines themselves can cause physical illness. What are those conditions? One of them is chronic stress.

Your body also contains little packets of genetic material that sit on the ends of your chromosomes. They’re called telomeres. Each time your cells divide, their telomeres get a little shorter, and when they become too short, you die. This is normal aging. But guess what else shrinks your telomeres?  Chronic stress .

If words can cause stress, and if prolonged stress can cause physical harm, then it seems that speech — at least certain types of speech — can be a form of violence. But which types?

This question has taken on some urgency in the past few years, as professed defenders of social justice have clashed with professed defenders of free speech on college campuses. Student advocates have protested vigorously, even violently, against invited speakers whose views they consider not just offensive but harmful — hence the desire to silence, not debate, the speaker. “Trigger warnings” are based on a similar principle: that discussions of certain topics will trigger, or reproduce, past trauma — as opposed to merely challenging or discomfiting the student. The same goes for “microaggressions.”

This idea — that there is often no difference between speech and violence — has stuck many as a coddling or infantilizing of students, as well as a corrosive influence on the freedom of expression necessary for intellectual progress. It’s a safe bet that the  Pew survey data released on Monday, which showed that Republicans’ views of colleges and universities have taken a sharp negative turn since 2015, results in part from exasperation with the “speech equals violence” equation.

The scientific findings I described above provide empirical guidance for which kinds of controversial speech should and shouldn’t be acceptable on campus and in civil society. In short, the answer depends on whether the speech is abusive or merely offensive.

Offensiveness is not bad for your body and brain. Your nervous system evolved to withstand periodic bouts of stress, such as fleeing from a tiger, taking a punch or encountering an odious idea in a university lecture.

Entertaining someone else’s distasteful perspective can be educational. Early in my career, I taught a course that covered the eugenics movement, which advocated the selective breeding of humans. Eugenics, in its time, became a scientific justification for racism. To help my students understand this ugly part of scientific history, I assigned them to debate its pros and cons. The students refused. No one was willing to argue, even as part of a classroom exercise, that certain races were genetically superior to others.

So I enlisted an African-American faculty member in my department to argue in favor of eugenics while I argued against; halfway through the debate, we switched sides. We were modeling for the students a fundamental principle of a university education, as well as civil society: When you’re forced to engage a position you strongly disagree with, you learn something about the other perspective as well as your own. The process feels unpleasant, but it’s a good kind of stress — temporary and not harmful to your body — and you reap the longer-term benefits of learning.

What’s bad for your nervous system, in contrast, are long stretches of simmering stress. If you spend a lot of time in a harsh environment worrying about your safety, that’s the kind of stress that brings on illness and remodels your brain. That’s also true of a political climate in which groups of people endlessly hurl hateful words at one another, and of  rampant bullying  in school or on social media. A culture of constant, casual brutality is toxic to the body, and we suffer for it.

That’s why it’s reasonable, scientifically speaking, not to allow a provocateur and hatemonger like Milo Yiannopoulos to speak at your school. He is part of something noxious, a campaign of abuse. There is nothing to be gained from debating him, for debate is not what he is offering.

On the other hand, when the political scientist Charles Murray argues that genetic factors help account for racial disparities in I.Q. scores, you might find his view to be repugnant and misguided, but it’s only offensive. It is offered as a scholarly hypothesis to be debated, not thrown like a grenade. There is a difference between permitting a culture of casual brutality and entertaining an opinion you strongly oppose. The former is a danger to a civil society (and to our health); the latter is the lifeblood of democracy.

By all means, we should have open conversations and vigorous debate about controversial or offensive topics. But we must also halt speech that bullies and torments. From the perspective of our brain cells, the latter is literally a form of violence.

Lisa Feldman Barrett, PhD  is a University Distinguished Professor at Northeastern University, and a member of the CLBB Scientific Faculty. 

Read the full article, originally published in  The New York Times.

• Category: News
• Tags: emotion | Lisa Feldman Barrett | speech | stress
• Date: July 14, 2017
• Author: The New York Times

• Subscribe to our feed
• Cases of Interest
• Publications
• Uncategorized
• February 2024
• November 2023
• February 2023
• November 2022
• October 2022
• February 2022
• January 2022
• October 2021
• December 2020
• November 2020
• September 2020
• August 2020
• February 2020
• January 2020
• August 2019
• February 2018
• January 2018
• December 2017
• August 2017
• February 2017
• January 2017
• December 2016
• November 2016
• October 2016
• September 2016
• February 2016
• January 2016
• December 2015
• November 2015
• October 2015
• September 2015
• August 2015
• February 2015
• January 2015
• December 2014
• November 2014
• October 2014
• September 2014
• August 2014
• February 2014
• January 2014
• December 2013
• November 2013
• October 2013
• September 2013
• August 2013
• February 2013
• January 2013
• December 2012
• October 2012
• September 2012
• February 2012
• December 2011
• September 2011
• December 2009
• October 2009
• October 1993
• Justin Baker
• Judith Edersheim
• Bruce Price
• Joshua Buckholtz
• Edward Hundert
• Nancy Gertner
• Jeffrey Rosen
• Daniel L. Schacter
• Ekaterina Pivovarova
• Justin Stahl
• Eric Huttenbach
• Lisa Feldman Barrett

NASP: The National Association of School Psychologists

• NASP Condemns Hate Speech and Violence, Calls on Schools to Reinforce Safe, Supportive School Environments for All Students

In This Section

• Senators Introduce Bipartisan, NASP-Led “CARE for Student Mental Health Act”
• National Association of School Psychologists Announces Winners of Annual Awards
• NASP Condemns Antisemitism and Islamophobia Amidst Violence in Israel and Gaza
• NASP Applauds Announcement of White House Office to Prevent Gun Violence
• NASP’s Executive Director, Kathleen Minke, Announces Retirement
• NASP Opposes Drastic Cuts in House FY24 Education Spending Bill
• NASP Offers Resources Related to the One-Year Anniversary of the Uvalde Tragedy
• NASP Applauds Bicameral, Bipartisan Introduction of Bill to Address Critical Shortages of School Psychologists
• NASP Calls for Reinstatement of the Assault Weapons Ban and Other Gun Safety Laws
• NASP Applauds Investments in President Biden’s FY2024 Budget Request
• NASP Opposes Censorship in Public Education
• Meaghan Guiney Named New Editor of Communiqué
• NASP Announces 2023 School Psychologist of the Year and Lifetime Achievement Award Recipients
• NASP Applauds Bipartisan Safer Communities Act
• NASP Adopts Resolution Regarding Policing in Schools
• NASP Calls on Congress to Pass Gun Safety Legislation
• NASP Urges Support for Psychological Safety in Schools, Calls for Action to Reduce Gun Violence
• NASP Obtains Council for Higher Education Accreditation (CHEA) Recognition as an Accrediting Organization
• Letter from President Laurie Klose Regarding NASP’s Position on the Use of Police in Schools
• NASP Supports Access to an Affirming School Environment and Uncensored Public Education
• NASP Welcomes Robust Investments in President Biden’s FY2023 Budget Proposal
• NASP Applauds Investments in FY22 Spending Package
• NASP Condemns State and Local Efforts to Undermine Transgender Youth Rights and Well-Being
• NASP Annual Convention
• April Turner Receives NASP 2022 School Psychologist of the Year Award
• NASP Urges Swift Passage of FY22 Federal Education Spending Bills
• NASP Applauds Biden Administration for Civil Rights Protections for Transgender Students
• President’s Proposed Budget Represents Historic Investments in Public Education System
• School Safety Experts Release Updated Guidance on Conducting Armed Assailant Drills in Schools
• NASP Guidance for Ensuring Student Well-Being in the Context of the Chauvin Trial
• NASP Adopts Resolution Supporting Efforts to Reduce the Harmful Effects of Climate Change on Children and Youth
• NASP Condemns Violence Against AAPI Communities, Urges Schools to Reinforce Students’ Safety, Well-Being
• NASP Announces 2021 School Psychologist of the Year and Lifetime Achievement Award Recipients
• NASP Statement on the Assault on the U.S. Capitol
• NASP Adopts Resolution Committing to Antiracism Action
• NASP Congratulates President-Elect Biden and Vice President-Elect Harris
• NASP Guidance for Ensuring Student Well-Being in the Context of the 2020 Election
• NASP Joins Colleague Organizations in Clarifying Use of SROs in Schools
• National Organizations Call for Rigorous Training and Appropriate Use of School Resource Officers
• NASP Statement on Reopening Schools for In-Person Learning
• NASP Urges Swift Passage of FY21 Federal Education Spending Bills
• Espinoza Decision Undercuts Public Education and Expands Taxpayer-Funded Discrimination
• NASP Calls for Action to End Racism and Violence Against People of Color
• President’s Proposed Budget Undermines Public Education System
• NASP Partners With Routledge, Taylor & Francis Group to Move School Psychology Review to New Online Platform
• Shane Jimerson Named New Editor of School Psychology Review
• NASP Condemns Hate-Driven Violence, Urges Schools to Reinforce Students’ Safety, Well-Being
• NASP Applauds Bill to Increase Access to Mental Health Services in Schools
• NASP Calls for Immediate Improvement in the Conditions and Care Provided Children in U.S. Custody; Endorses Open School Psychology Letter
• NASP Urges Support for the Increasing Access to Mental Health in Schools Act
• NASP Urges Continued Caution in Media Coverage of School Shootings
• President’s Proposed Budget Presents Challenges for Public Education System
• Parkland and Other Anniversaries Offer Opportunity to Highlight Effective School Safety Efforts
• NASP Expresses Concern Regarding Federal School Safety Report
• NASP Affirms Support for Civil Rights Protections of Transgender People in Title IX
• New Research Highlights Approaches to Help Close the Discipline Gap in U.S. Schools
• NASP Opposes Federal Funds to Support Arming Teachers
• NASP Leader Addresses the Effect of Media Coverage on School Violence at Federal Safety Commission Meeting
• Leading Education Organizations Release Guidance to Safe Schools Framework (PDF)
• NASP Calls for End to Policy Separating Families at the Border
• NASP Calls for an Increase in Access to Suicide Prevention Resources
• Preventing Mass Violence Requires Access to Mental Health Services and Reduced Inappropriate Access to Firearms - Santa Fe, Texas
• Guidance Regarding 13 Reasons Why, Season 2
• Kathleen Minke Named as New NASP Executive Director
• Omnibus Spending Bill Provides Increased Investments for School Safety and Mental Health Services
• NASP Opposes Arming Teachers
• Preventing Mass Violence Requires Access to Mental Health Services and Reduced Inappropriate Access to Firearms
• Responsible Media Coverage of Crisis Events Impacting Children and Youth
• President's Proposed Budget Presents Challenges for the Well-Being and Education of Millions of Children and Youth
• NASP Adopts Resolution Supporting Efforts to Prevent Gun Violence
• NASP Calls for Comprehensive School Safety Measures and Common Sense Gun Violence Prevention Efforts
• NASP Awards Lifetime Honorary Membership to Crockett, Gorin, and Reynolds
• NASP Executive Director Susan Gorin Announces Retirement After 25 Years
• President’s Proposed Budget Jeopardizes the Well-Being and Education of Millions of Children and Youth
• NASP Statement Regarding Terrorist Attack in Manchester, England
• NASP Adopts Resolution Affirming Commitment to High-Quality Public Education for All Children and Youth
• NASP Strongly Disagrees With the Trump Administration’s Decision to Rescind Federal Guidance Related to Civil Rights Protections for Transgender Students
• NASP Passes Resolution Reaffirming Commitment to Promoting the Rights of All Students
• NASP Guidance for Reinforcing Safe, Supportive and Positive School Environments for All Students
• NASP Commends Hillary Clinton’s New Bullying Prevention Plan
• NASP Applauds Report Calling for Increased Access to School Psychologists
• NASP Statement Regarding Recent Acts of Violence
• 5,000 School Psychologists Gather to Promote Successful & Mentally Healthy Students
• NASP Applauds President Obama's Efforts to Reduce Violence and Improve Gun Safety
• NASP Applauds Passage of ESSA
• NASP Urges Congress to Pass ESSA
• Paris Statement
• NCSP Earns Formal Recognition From National Register
• Todd A. Savage to Begin Term as NASP President
• Joe Prus Named NASP 2015 Lifetime Achievement Award Recipient
• Pamela Agan-Smith Receives NASP 2015 School Psychologist of the Year Award
• NASP and NASRO Release Guidance on Conducting Active Shooter Drills in Schools
• Best Practice Related to SLD Identification
• Dr. Sylvia Rosenfield Named NASP 2013 Lifetime Achievement Award Recipient
• Educators Call for Interdisciplinary, Multitiered Framework to Improve School Safety & Access to Mental Health Supports for Students
• NASP Cautions Against Arming School Personnel to Improve School Safety
• NASP Recommendations for Comprehensive School Safety Policies
• NASP Partners With Lady Gaga for Born This Way Bus Tour
• Stigmatizing Mental Health or Developmental Disorders is Harmful
• NASP Sandy Hook Statement
• NASP and NASRO Urge Accuracy With Regard to Lockdown Versus Other Types of Armed Assailant Drills

Bethesda, MD—The past two weeks in the United States have been fraught with serious incidents of hate and violence. The National Association of School Psychologists (NASP) strongly condemns all hate speech and crimes, racist behavior, antisemitism, misogyny, xenophobia and any behavior that undermines our collective sense of decency and security as a nation and threatens the sense of personal safety for millions of people in targeted populations across the country. Our deepest sympathies are with all those directly affected by the recent horrific shootings in Jeffersontown, KY, Pittsburgh, PA, and Tallahassee, FL.

Such hate and acts of violence must stop.

As a nation, we have much to do to address the underlying causes of hate-based behavior, including changing the tone of our national discourse and enforcing civil rights and other legal protections. As parents, caregivers, and educators, we have a critical responsibility to help children and youth feel safe and secure and learn how to engage with others of differing viewpoints in a peaceful, tolerant, and respectful manner.

Schools are essential to this process. It is imperative that educators create positive school communities in which violence is not tolerated; people at risk are identified and helped; inequity is addressed; problem solving, rather than blame and disregard for challenges, is the norm; and people of all backgrounds, races, religions, and cultures are valued and engaged as equals. Specific recommendations include:

Reinforce a sense of positive school community. Establishing positive relationships between adults and students is foundational to safe, successful learning environments. Such relationships are built on a sense of mutual trust and respect. Maintain culturally and linguistically responsive practices and ensure that students and their families feel connected and engaged. We function as a society only when we have that shared sense of relationship; helping children identify and develop those relationships is vital.

Reassure children that adults will take care of them. Many children and youth may feel at risk. It is important to reinforce strategies to ensure both physical and psychological safety. Remind adults and students of the importance of supporting each other during difficult times and acknowledge people will have a variety of emotions. If students feel physically or psychologically unsafe, they need to know how to report incidents, and trusted adults must be there to validate and respond to their concerns.

Model and teach desired behaviors. We know that adult actions and attitudes influence children. Adults can help children and youth manage their reactions to events in the news and their communities by understanding their feelings, modeling healthy coping strategies, and closely monitoring their own emotional states and that of those in their care. Identifying and redirecting negative thoughts and feelings can help to teach children social–emotional skills and problem solving. Adults should never engage in mocking, belittling or threatening behavior.

Help children manage strong emotions. For many children and youth, incidents of violence, media images, and messages to which they are exposed can trigger a range of strong emotions. Some children may experience anger or stress. Others may feel a sense of fear. Children’s emotions often spill over into schools. Help children understand the range of emotions that they are feeling and to learn to express them in appropriate and respectful ways. For children experiencing stress, we can help by spending time with them, encouraging them to talk about their feelings, maintaining a sense of normalcy in their schedules and activities, and teaching coping strategies.

Reinforce acceptance and appreciation for diversity as critical American values. Acknowledge that everyone is entitled to their personal opinions but that hateful or intolerant comments about others’ cultures, sexual orientations, religions, or races—or any other comments that are meant to hurt or make another feel threatened, unsafe, or unwelcome—will not be tolerated. This includes adult behavior.

Stop any type of harassment or bullying immediately. Make it clear that such behavior is unacceptable. Talk to the children involved about the reasons for their behavior. Offer alternative methods of expressing their anger, confusion, or insecurity, and provide supports for those who are subject to bullying. School staff should encourage students to continue to be respectful of others.

Encourage children to channel their views and feelings into positive action. Like adults, children and youth are empowered by the ability to do the right thing and help others. Working with classmates or members of the community who come from different backgrounds not only enables children to feel that they are making a positive contribution, it also reinforces their sense of commonality with diverse people.

School psychologists play a critical role in helping schools create, supportive learning environments. They work with school staff and families to establish positive school climates, prevent bullying, harassment and violence, establish equitable and culturally-responsive policies and practices, and support students’ mental health. Related NASP resources include:

• Bullying Prevention & Intervention in Schools
• Racism, Prejudice, & Discrimination
• Safe and Supportive Schools for Lesbian, Gay, Bisexual, Transgender, Questioning (LGBTQ+)
• Youth Resolution: Affirming the Rights to Safe and Supportive Schools and Communities for All Students

The National Association of School Psychologists (NASP) is a professional association that represents more than 25,000 school psychologists. The world's largest organization of school psychologists, NASP works to advance effective practices to improve students' learning, behavior, and mental health. Our vision is that all children and youth thrive in school, at home, and throughout life.

• Open access
• Published: 04 October 2018

Emotion recognition and school violence detection from children speech

• Tian Han   ORCID: orcid.org/0000-0002-4385-6082 1 , 2 ,
• Jincheng Zhang 1 ,
• Zhu Zhang 2 , 3 ,
• Guobing Sun 2 , 6 ,
• Liang Ye 2 ,
• Hany Ferdinando 2 , 7 ,
• Esko Alasaarela 2 ,
• Tapio Seppänen 5 ,
• Xiaoyang Yu 4 &
• Shuchang Yang 1  

EURASIP Journal on Wireless Communications and Networking volume  2018 , Article number:  235 ( 2018 ) Cite this article

3126 Accesses

7 Citations

15 Altmetric

Metrics details

School violence is a serious problem all over the world, and violence detection is significant to protect juveniles. School violence can be detected from the biological signals of victims, and emotion recognition is an important way to detect violence events. In this research, a violence simulation experiment was designed and performed for school violence detection system. Emotional voice from the experiment was extracted and analyzed. Consecutive elimination process (CEP) algorithm was proposed for emotion recognition in this paper. After parameters optimization, SVM was chosen as the classifier and the algorithm was validated by Berlin database which is an emotional speech database of adults, and the mean accuracy for seven emotions was 79.05%. The emotional speech database of children extracted in violence simulation was also classified by SVM classifier with proposed CEP algorithm, and the mean accuracy was 66.13%. The results showed that high classification performance could be achieved with the CEP algorithm. The classification result was also compared with database of adults, and the results indicated that children and adults’ voice should be treated differently in speech emotion recognition researches. The accuracy of children database is lower than adult database; the accuracy of violence detection will be improved by other signals in the system.

1 Introduction

School violence happens in school all over the world. It is recognized as one of the main reasons for dropping out of school, adolescent suicide, and even one of the causes leading to crime of youth including school shooting. Violence events in school can be stopped by classmates, teachers, or parents, but many of them do not know their friends, pupils, or children are victim of school violence events [ 1 , 2 ].

Some violence detection system is designed to help the bullied children. The ICE Blackbox is a personal security app. When a violence event occurs, the user is able to press a button to activate the ICE Blackbox. Then, the system will record the audio, video, and GPS location and send the information to the ICE Blackbox secure servers. In the app Tip Off, school violence as well as crimes can be reported to the server by text. So far, this kind of apps all requires manual activation, but most of the time, it is difficult for bullied children to active the app manually in real violence situation. So it is urgent to develop a system that is able to detect the status of children and send information and alarm automatically when violence happens [ 3 , 4 ].

Physical and psychological traumas are the two main injuries caused by school violence. By using sensors and pattern recognition technology, the status of human body and emotion can be detected. This technique has proved very useful for identifying violence because the changes of human body and emotion can be detected during violence events. Physical movement changing such as being hit, being pushed, and falling down which occur frequently when being bullied can be detected by accelerometer and gyroscope. Some negative emotions, such as anger, fear, anxiety, and sadness, can be detected from voice signal, ECG signal, and so on. Violence events can be detected, and also, the accuracy will be improved by combining emotion and physical changing detection [ 5 , 6 ]. The goal of our research is to design a violence detection system using these signals which can detect violence events and notify the responsible person automatically to stop the violence events.

This paper focuses on emotion recognition of voice signal in the violence detection system. In Section 2 , emotions in violence events are analyzed. In Section 3 , a kind of speech emotion recognition algorithm with consequential eliminating process (CEP) is presented. Then, parameters of support vector machine (SVM) are optimized, and Berlin voice database is used to test and verify the algorithm. In Section 4 , a school violence simulation experiment performed in an elementary school is described, from which voice signals database are extracted. The database extracted from the experiment is processed by CEP algorithm, and the calculation result is presented and analyzed.

2 Emotions in violence events

What kind of emotions pupils have during the violence events is an important issue for emotion detection research. A research focused on emotional reaction of school violence victims has carried on 6282 Maltese schoolchildren between 9 and 14 years of age. The results showed that most pupil victims felt angry, vengeful, helpless, and self-pity, and about 24% of the victims felt indifferent [ 7 , 8 ]. Another research presented a survey on violence victims in three countries (England, Italy, and Spain) revealed that the victims are angry (42.7%), upset (34.8%), stressed (22.4%), worried (24.3%), afraid (18.1%), alone (14.3%), defenseless (14.3%), and depressed (18.9%) [ 9 ]. Both the researches showed that the victims have negative emotions during the violence events.

Generally, the term emotion describes the subjective feelings in short periods of time which are related to events, persons, or objects [ 10 , 11 ]. Since the emotional state of human is a highly subjective experience, it is hard to find objective and universal definitions. This is the reason there are different approaches to model emotions in the psychological literature. One approach is the definition of discrete emotion classes, the so-called basic emotions. Ekman defined seven emotions which humans are very familiar with: happiness, sadness, anger, anxiety, boredom, disgust, and neutral [ 12 ]. These seven emotions are considered as the basic emotions, and more emotions can be defined by mixtures of the basic emotions [ 13 ]. According to this theory, the emotions, which the school violence victims have in above two researches, can also be defined by mixtures of the basic emotions. Analyzing these emotions generated by school violence victims, the negative emotions consist of three basic emotions: anger, sadness, and fear. So the detection of these three basic emotions may indicate that violence events happened. Combined with movement and other bio-signal detection, the accuracy of violence detection might be improved.

3.1 Classification algorithm with consecutive elimination process

In pattern recognition researches, feature is the most important parameter to distinguish one kind of speech from another. Since many features can be extracted from voice signals, it is critical that right features are selected to classify different labels. Feature selection is often used to choose the best feature set to classify the labels. It appears that when classifying more labels, it is challenging to get higher accuracy. The reason is that in emotion recognition, different emotion is sensitive to different features. When a useless feature is added to a good set of features, the performance of the classifier will decrease. It was found that feature selection gives the highest accuracy when the number of labels is two. So when emotional speeches are classified, it is better to make the classifier always working between two labels. In this paper, consecutive elimination process (CEP) is proposed to classify the emotional speech. It makes classifiers working between two labels, and the best feature set is chosen for each classifier in order to get the highest accuracy.

Before the CEP, feature selection work is done between each two labels within the database. The best feature set and the classify accuracy of each two labels can be get through this procedure. A table with labels, selected feature sets, and classifier accuracy of each two labels which is called LFA table is created.

Figure  1 a shows an example of this process. Suppose the database has n classes of objects and the labels are marked as 1, 2, 3, …, n . For this n -label database, n ( n  − 1)/2 times of feature selection is done between each two labels. The LFA table contains the results of each feature selection procedure. The first column of the LFA table records the number of labels such as (1, 2), (1, 3), (1, 4), …, ( n  − 1, n ). The following columns are the corresponding classifier (C1, C2, C3,…, C k ), the selected feature set (Fset1, Fset2, Fset3, …, Fset n ), and the classify accuracy (A1, A2, A3, …, A k ) in which k  =  n ( n  − 1)/2. With this LFA table, k kinds of classifiers can be set up with corresponding feature set to classify any two labels in the database.

Example of consequential eliminating process in LFA table. Figure  1 shows a process of CEP algorithm. Suppose A i is the maximum accuracy value in a . For the first layer, the classifier C i with the maximum accuracy is chosen. The testing sample is then classified by classifier C i with feature set Fset i between label 2 and label n . Suppose the result of the first layer is n . It is believed that the testing sample does not belong to label 2. Then, lines with label 2 are eliminated from the original LFA table as shown in b . For the second layer, the above processes are repeated again with the new LFA table shown in c and another label is eliminated from the new LFA table. The rest may be deduced by analogy. After n  − 2 layers of classification, there will be only one line left in the LFA table. This last classifier gives the final predicted label of the testing sample

The CEP begins when a testing sample is ready to be classified. This process divides the classification procedure into several layers. In each layer, one possibility of predicted label is excluded until the final predicted label is calculated out. And for each layer, the classifier with the highest accuracy in LFA table is chosen.

Take Fig.  1 as an example. Suppose A i is the maximum accuracy value in Fig.  1 a. For the first layer, the classifier C i with the maximum accuracy is chosen. The testing sample is then classified by classifier C i with feature set Fset i between label 2 and label n . Suppose the result of the first layer is label n . Since there are n labels in the database and only two labels are used in the first layer classifier, the result label n does not mean that the final predicted result of the testing sample is label n . While it is believed that the testing sample does not belong to label 2 because label 2 is excluded during the first layer classifier, then lines with label 2 are eliminated from the original LFA table as shown in Fig.  1 b. For the second layer, the above processes are repeated again with the new LFA table shown in Fig.  1 c and another label is eliminated from the new LFA table. The rest may be deduced by analogy. After n  − 2 layers of classification, there will be only one line left in the LFA table. This last classifier gives the final predicted label of the testing sample.

In the CEP, a high classify accuracy is guaranteed because the classifier with the highest accuracy is chosen in each layer.

3.2 SVM and parameters optimization

SVM is a classifier widely used in classification of two classes. Theoretically, SVM is divided by hyperplane and hyperplane is the decision boundary between two classes [ 14 ]. Figure  2 is an example of two classes classification, and the hyperplane is a line in two-dimensional space (green circle and red X are different classes of data).

The principle of SVM hyperplane in two-dimensional space. The green circle and red X stand for two kinds of data. A line separates two kinds of data in the space which is the hyperplane in the space

If there are many hyperplanes that can divide the data as shown in Fig.  3 , a best one should be chosen. It is considered that the distance between hyperplane and the data nearest to the hyperplane should be as far as possible, because the data is further from the boundary and the probability of error is smaller.

Shows the principle of best hyperplanes selection. In two-dimensional space, the training data (green circle and red X) can be separated by many hyperplanes, so selection is an important process for the classifier. That is the principle of parameters optimization of SVM

So optimizing SVM is to choose the best hyperplane. Optimization strategy is to make the distance between the hyperplane and the nearest data farther. Therefore, SVM is called “the large margin classifier.” The question can be described by the following mathematical formulas.

‖ ω ‖ is the bound norm of ω . γ i is the distance between nearest data and the hyperplane, and it is usually supposed that γ i  = 1 to simplify the calculation.

In machine learning, convex optimization method is often used to solve optimization question. Therefore, the mathematical formulas can be transformed from \( \max \frac{1}{\left\Vert \omega \right\Vert } \) to

Lagrange duality is efficient to solve the problem, because it is easy to solve and bring in kernel function to solve nonlinear problems. Lagrange function is established as below.

Derivation of the function presents,

Sequential minimal optimization is used to calculate b and α . Through the above method, strict equation of hyperplane is got, which allows little error. Soft margin allows some errors in the classification, and the error is controlled not to be large; therefore, a penalty coefficient is added. The optimized model changes.

C is a penalty coefficient of the soft margin. The optimizing method which is the same as the model does not have a soft margin.

The core of SVM is the kernel functions. Kernel function converts computation from high-dimensional space to low-dimensional space, that means it can map after calculating. This helps us to reduce a large amount of calculation and save calculating time. Parameter g is in the kernel function and should be optimized for better performance.

Parameters optimization is an important process. Parameters of SVM affect the accuracy of the classification results. The classification algorithm proposed in part 3 uses SVM classifier between each two labels, and parameters optimization is needed for every classifier, so n ( n  − 1)/2 sets of parameters should be optimized if there are n labels in the database.

K-fold cross-validation is used in the parameters optimization process. Firstly, the database is divided into two parts: 33% of which is used for testing data and 67% of which is used for training data, and secondly, the training data is divided into two parts: 25% of which is used for training SVM and 75% of which is used for validation of parameters optimization, as shown in Fig.  4 . For cross-validation divides the database randomly, parameters optimization works four times and makes sure the whole training data can be used as validation. The mean accuracy is used to determine the optimized parameters. After parameters optimization, the testing data is used to compare the accuracy between default parameters and optimized parameters. This will increase the adaptation of the classifier, and the result is more convictive because the testing data is not used during the parameters optimization process at all.

Shows the data validation process. The database is divided into two parts. Firstly, 33% of which is used for testing data and 67% of which is used for training data. Secondly, the training data is divided into two parts, 25% of which is used for training SVM and 75% of which is used for validation of parameters optimization

For SVM, classifier parameters c and g are needed to be optimized. The range of c is from 0.1 to 10 with step length of 0.1, and the range of g is from 0.01 to 1 with step length of 0.01 parameters optimization.

In order to verify the algorithm, the Berlin database with emotional speech, which is very popular in emotion recognition research, is analyzed by this algorithm. Berlin database contains 535 sentences spoken by 10 actors in happy, angry, fearful, sad, bored, disgusted, and neutral version. These seven emotions are marked from 1 to 7 as seven labels shown in Table  1 . In the following content, labels 1 to 7 are used instead of the seven emotions. The analyzing results of Berlin database can be easily compared with other research. If the accuracy of the algorithm proposed above is good, the algorithm is effective in emotion recognition and can be used to analyze the data in violence simulation.

Berlin database is used to optimize the parameters. There are seven emotions in the database, so 21 sets of parameters should be optimized. The optimized parameters are shown in Table  2 . For different labels, the optimized parameters are different and the classification accuracy increases after parameters optimization.

3.3 Algorithm verification by Berlin database

3.3.1 feature extraction.

The first step to deal with the database is feature extraction. The emotional speech is usually divided into frames by Hamming window, and original features are extracted from the emotional speech by frames. In this research, the following original features are used which are shown in Table  3 , and they are marked as f1 to f16 in the following content.

The software “Opensmile” is used to extract the features from speeches in the Berlin database. There are 384 features as statistical functional is applied to low-level descriptor contours. The contour is smoothed after extracting the original value by frame. The smoothing method “sma” indicates that they were smoothed by a moving average filter with window length 3. And the smoothing method “de” indicates that the current feature is a first-order delta coefficient (differential) of the smoothed low-level descriptor. Then, the statistical function is applied to the contour, and 12 statistical functions which are marked as S1 to S12 are presented in Table  4 .

Figure  5 shows the detailed feature process procedures. The original feature contours (f1–f16) are first smoothed by two methods separately, and then 12 statistic functions are applied to the smoothed contours. Figure  5 also shows the numbering scheme for the 384 features. Final features got from f1–f16 with statistic functions S1–S12 and smoothing method “sma” are numbered as F1–F192. And final features got from smoothing method “de” are numbered as F193–F384. With this numbering scheme, the source of the final feature can be found out easily and it is convenient to use the numbered features in the following work.

Shows the detailed feature process procedures. The original feature contours (f1–f16) are first smoothed by two methods separately and then 12 statistic functions are applied to the smoothed contours. Figure  5 also shows the numbering scheme for the 384 features

3.3.2 Features selection

Feature selection is an important step before a classifier is set up. With the proper feature set, the accuracy of the classifier would be higher. And even with different training and testing samples in the same database, the result of feature selection process for the same classifier would be different.

Sequential forward-floating search method is used between each two labels in the Berlin database. Since with different samples the feature selection result would be different, the feature selection process was run 100 times for each two labels. The set with the maximum number of repetitions in the 100 results was chosen as the final selected feature set for the corresponding two labels. As SVM classifier would be used in the CEP procedure, the feature selection process also used SVM to generate the discriminant function.

Table  5 is the LFA table got from the Berlin database using the above feature selection method. The second column is the final features number, which can be found above in Fig.  5 .

3.3.3 Calculation results

The Berlin database is classified with CEP procedure using LFA table in Table  5 . There are two sets of labels with the maximum accuracy of “100%” in the LFA table, and the first layer of classifier is chosen randomly from them. Fourfold cross-validation is used to divide the database into training data and testing data, and SVM classifier with optimized parameters is used. The classify process is run 25 times to reduce contingency because the fourfold cross-validation divides the database randomly.

The results show that with the CEP procedure, the highest accuracy is 82.24%, the lowest accuracy is 76.44%, the mean accuracy is 79.05%, and the standard deviation is 0.0147. The confusion matrixes of the best result are shown in Table  6 in detail.

Berlin database is often used in verification and comparison of algorithm for speech emotion recognition. The results with CEP procedure show a high classify accuracy, which indicates that CEP procedure works well for speech emotion recognition. So this method can be used to analyze the emotional speech extracted from the violence simulation experiment in the following work.

4 Experiments

4.1 violence simulation experiments.

A violence simulation was conducted to collect signals for this study at Normaalikoulu Elementary School in Oulu, Finland. The experiments were designed by some psychologists in University of Oulu in Finland. The second- and sixth-grade pupils of Normaalikoulu Elementary School in the city of Oulu, Finland, joined the experiments, and they were arranged in three-pupil small groups in the classroom. The pupils take turns to play as bullies and victims. The drama series were used to create school violence activities which are used to simulate the real school violence. Of course, the experiments are allowed by the pupils’ parents, and the detailed experiment plan was discussed and permitted by the Ethics Committee of University of Oulu, so that the experiment will not affect the pupils anyway. The whole experiments are divided into five parts as follow.

The first task of every group was to simulate verbal violence (being called names or insulted) to one member of the group in the restricted area 3 × 3 m. The group was not allowed to step out of the area; otherwise, the trial was stopped and started again. Each member of the group was simulated bullied in random order by the other two members of the group, and physical contact was forbidden.

The second task is freedom physical violence game. One pupil played the role of the victim and was asked to try his/her best to stay in the 3 × 3 m 2 . Two pupils played the role of bullies and tried to push the victim out of the area. When the victim was pushed out, the game stopped. The members played the role of victim in turn.

The third task is emotional speech. All the pupils were taught and led to speak some sentences in five different basic emotions. Non-emotional sentences such as “There is a green carpet on the floor.” are chosen in this part. The five basic emotions were happiness, sadness, fear, anger, and neutral.

The fourth task is special activities and movements simulation. Some special activities happened during violence events were simulated, such as being pushed from various direction, being stumbled down, being knocked with shoulder, and being shaken by holding the shoulders. These activities were carried out under protection of soft mattresses. And also, some common movements, such as walking, running, jumping, and playing normal games, are carried out control.

The last part is relaxation. All the pupils lay down on the mattresses and close their eyes. One psychology teacher told a relaxing story with light music. The purpose of this section was to make both children’s physical and mental state restored to normal.

Data was collected from pupils using different procedures including pre- and post-interviews and video analysis. The activities of the groups were video recorded by MORE recording system (Multimodal Recording System) for the later analysis [ 15 ]. The emotional speeches in task three are also recorded by MORE recording system. The heart rate variability (HRV) was measured using beat-to-beat RR-intervals with Zephyr heart rate monitoring system (Zephyr Co.). The HRV recordings were synchronized with MORE recording system. The postures, breathing rates, body temperatures, saliva samples, and speech signals were also collected. In addition, acceleration sensors were used to measure participants’ physical movements during the tasks, and with the aid of the developed algorithms, the violence events were automatically collected from the raw data for further analysis. Open-ended interview questions were used in pre- and post-interviews to define the mood and emotions of the participants.

The emotional speeches recorded in task three were extracted and analyzed in this paper. It was believed that during verbal and physical violence simulation in tasks 1 and 2, negative emotions would be generated. The following emotional speech task would record part of the real negative emotion generated during violence simulation so that analyzing the emotional speeches in task 3 was approximately equal to analyzing the voice signals during violence situation.

4.2 Children database

The emotional speeches of children are extracted from the emotional speech task of the violence simulation experiments. Children are led to act a violence simulation event designed by psychologists. It is believed that negative emotions will generate during this simulation, which is proved by the feeling meter test of the children before and after the violence simulation. The emotional speech task is performed just after the simulation so part of the negative emotions will be kept during the emotional speech task.

In the emotional speech task, totally, 12 children are asked to speak three sentences in five emotions: happiness, sadness, fear, anger, and neutral. Three girls and three boys of the children are from grade 2, and another three girls and three boys are from grade 6. The three sentences are in Finnish that are shown in Table  7 , because the experiment is performed in Finland. The children are asked to speak each sentence in each emotion for two or three times. After getting rid of the voice in poor quality, there are 382 clips of speech in five emotions.

4.3 Classification of emotional speech extracted from violence experiment

The CEP procedure of classification is used to the database extracted from the violence experiment. The “Opensmile” is also used for feature extraction, and 384 features are extracted for each speech clip. These 384 features are the same as that in Section 5 and numbered also in the same way. Table  8 is the LFA table of the violence simulation database after feature selection processes.

Four cross-validations and SVM classifier are used for the classification, and each procedure is run for 25 times. The results show that the highest accuracy is 60.63%, the lowest accuracy is 53.68%, and the mean accuracy is 66.13%. The standard deviation is 0.0236. The confusion matrixes of the best result are shown in Table  9 in detail.

5 Results and discussion

Comparing the results from Berlin and violence simulation databases, it can be seen that the mean accuracy of the Berlin database is about 12% higher than the accuracy of violence simulation database from children and the range of the accuracy difference between max and min of violence simulation database is bigger than that of the Berlin database. The reason is that some children were always in a happy and excited state during the experiment, and they cannot control the voice as well as the adult actors in the Berlin database. In some speech in sad emotion, light laughing can be heard in the violence simulation database. That is the limitation of the violence simulation database.

From the LFA table, it is presented that the selected feature sets for the same two labels are totally different between adults and children. It indicates that children’s voice is different from adults, so the classification of adults and children’s speech should be treated differently in research of emotion recognition.

Also seen from the LFA table, for Berlin database, the two labels with the lowest classification accuracy are labels 1 and 2. It means that for adults, the emotions which are the most difficult to separate is happiness and anger. While for children’s speech in violence simulation database, the labels most difficult to separate is sadness and fear (lowest accuracy in LFA table). This is another proof of the difference between voices of children and adults, and it also shows the significance of classifying the children’s speech individually. Fortunately, sadness and fear are all emotions generated in violence events and the influence to violence detection system is not serious.

So far, the accuracy of reorganizing emotion from voice of children is not as high as adults, but this result is helpful for the violence events detection system. Combined with other signal such as ECG, movement, temperature, and breathing, the system may give good performance to detect school violence events and help to protect children in our future work.

6 Conclusion

In order to develop a school violence detection system, a school violence simulation experiment is proposed and the experiment is performed in Normaalikoulu Elementary School in Oulu, Finland. The emotional speech extracted from the experiment is analyzed in this paper.

A CEP procedure is proposed in this paper for the emotion recognition. Both Berlin and violence experiment databases are analyzed with the procedure using SVM classifier, and the mean accuracy for the two databases is 79.05% and 66.13%. It shows that the CEP procedure got a high accuracy for speech emotion recognition. Comparing Berlin and violence experiment databases’ result, the properly selected features and the difficulties of emotional speech recognition are different between adults and children. It indicates that children’s voice is different from adults, so the classification of adults and children’s speech should be treated differently in research of emotion recognition.

Though the accuracy of emotion recognition with CEP procedure for children database is not as high as that of adults’ database, the proposed experiment and CEP procedure will contribute to the multi-signal system for violence events detection.

Abbreviations

Consecutive elimination process

Electrocardiogram

Global Positioning System

Heart rate variability

Label feature accuracy

Support vector machine

D. Olweus, School bullying: development and some important challenges. Annu. Rev. Clin. Psychol. 9 , 751–780 (2013). https://doi.org/10.1146/annurev-clinpsy-050212-185516

Article   Google Scholar  

E. Menesini, C. Salmivalli, Bullying in schools: the state of knowledge and effective interventions. Psychol. Health Med. 22 , 240–253 (2017). https://doi.org/10.1080/13548506.2017.1279740

J. Wang, R.J. Iannotti, T.R. Nansel, School bullying among adolescents in the United States: physical, verbal, relational, and cyber. J. Adolesc. Health 45 (4), 368–375 (2009). https://doi.org/10.1016/j.jadohealth.2009.03.021

T. Vaillancourt, R. Faris, F. Mishna, Cyberbullying in children and youth: implications for health and clinical practice. Can. J. Psychiatry 62 (6), 368–373 (2017). https://doi.org/10.1177/0706743716684791

A. Foteini, H. Dimitris, K. Anderson Adam, ECG pattern analysis for emotion detection. IEEE Trans. Affect. Comput. 3 (1), 102–115 (2012). https://doi.org/10.1109/T-AFFC.2011.28

L. Ye, H. Ferdinando, T. Seppanen, An Instance-Based Physical Violence Detection Algorithm for School Bullying Prevention (11th IEEE International Wireless Communications and Mobile Computing Conference, Dubrovnik, 2015), pp. 24–25

Google Scholar  

M.G. Borg, The extent and nature of bullying among primary and secondary schoolchildren. Educ. Res. 41 (2), 137–153 (1999). https://doi.org/10.1080/0013188990410202

L.R. Barhight, J.A. Hubbard, C.T. Hyde, Children’s physiological and emotional reactions to witnessing bullying predict bystander intervention. Child Dev. 84 (1), 375–380 (2013). https://doi.org/10.1111/j.1467-8624.2012.01839.x

M. Giménez Gualdo Ana, C. Hunter Simon, D. Kevin, The emotional impact of cyberbullying: differences in perceptions and experiences as a function of role. Comput. Educ. 82 , 228–235 (2015). https://doi.org/10.1016/j.compedu.2014.11.013

M.A. Quiros-Ramirez, T. Onisawa, Considering cross-cultural context in the automatic recognition of emotions. Int. J. Mach. Learn. Cybern. 6 (1), 119–127 (2015). https://doi.org/10.1007/s13042-013-0192-2

M. Waseem, M. Ryan, C.B. Foster, Assessment and management of bullied children in the emergency department. Pediatr. Emerg. Care 29 (3), 389–398 (2013). https://doi.org/10.1097/PEC.0b013e31828575d7

H. Saarimaki, A. Gotsopoulos, I.P. Jaaskelainen, Discrete neural signatures of basic emotions. Cereb. Cortex 26 (6), 2563–2573 (2016). https://doi.org/10.1093/cercor/bhv086

H. Schlosberg, Three dimensions of emotion. Psychol. Rev. 61 (2), 81–88 (1954). https://doi.org/10.1037/h0054570

X.D. Wu, V. Kumar, J.R. Quinlan, Top 10 algorithms in data mining. Knowl. Inf. Syst. 14 (1), 1–37 (2008). https://doi.org/10.1007/s10115-007-0114-2

A. Keskinarkaus, S. Huttunen, A. Siipo, MORE - a multimodal observation and analysis system for social interaction research. Multimed. Tools Appl. 75 (1), 6321–6345 (2016). https://doi.org/10.1007/s11042-015-2574-9

Download references

Acknowledgements

The authors acknowledge the Natural Science Foundation of Heilongjiang Province of China (Grant No. F201315) and Harbin Research Fund for Technological Innovation (No.2013RFQXJ104) and Scientific research project of Heilongjiang provincial department of education (No. 12541144).

The research presented in this paper was supported by the Heilongjiang Provincial Science and Technology Department of China, Heilongjiang Provincial Education Department of China, and Harbin Municipal Science and Technology Bureau of China.

Availability of data and materials

The data and materials in this paper are all true and available.

Author information

Authors and affiliations.

Department of Internet of Things Engineering, Harbin University of Science and Technology, Harbin, China

Tian Han, Jincheng Zhang & Shuchang Yang

Optoelectronics and Measurement Technique Unit, University of Oulu, Oulu, Finland

Tian Han, Zhu Zhang, Guobing Sun, Liang Ye, Hany Ferdinando & Esko Alasaarela

Department of Communication Engineering, Harbin University of Science and Technology, Harbin, China

Department of Measurement-control technology and instrumentation, Harbin University of Science and Technology, Harbin, China

Xiaoyang Yu

Physiological signal analysis team, University of Oulu, Oulu, Finland

Tapio Seppänen

Department of Automation, Heilongjiang University, Harbin, China

Guobing Sun

Department of Electrical Engineering, Petra Christian University, Surabaya, Indonesia

Hany Ferdinando

You can also search for this author in PubMed   Google Scholar

Contributions

TH is the main writer of this paper. He proposed the main idea of the algorithm for emotion recognition, joined the whole experiment, and calculated the results. JZ optimized the parameters of SVM. ZZ processed the children’s voice and founded the database. ZZ, HF, and GS all joined the experiment and helped extracted the data. EA and TS designed the experiment and helped finish the experiment. XY helped to improve the algorithm and the manuscript. SY wrote the code of classifier. All authors read and approved the final manuscript.

Corresponding author

Correspondence to Tian Han .

Ethics declarations

Ethics approval and consent to participate.

Not applicable.

Consent for publication

Competing interests.

The authors declare that they have no competing interests.

Publisher’s Note

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

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License ( http://creativecommons.org/licenses/by/4.0/ ), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.

Reprints and permissions

About this article

Cite this article.

Han, T., Zhang, J., Zhang, Z. et al. Emotion recognition and school violence detection from children speech. J Wireless Com Network 2018 , 235 (2018). https://doi.org/10.1186/s13638-018-1253-8

Download citation

Received : 13 June 2018

Accepted : 18 September 2018

Published : 04 October 2018

DOI : https://doi.org/10.1186/s13638-018-1253-8

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

• Emotion recognition
• Children speech
• Violence simulation

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
• Occup Ther Int
• v.2022; 2022

The Construction of an Action-Speech Feature-Based School Violence Recognition Algorithm and Occupational Therapy Education Model for Adolescents

Shuaiqing zhang.

1 Institute of Education, Joongbu University, Daejeon 32713, Republic of Korea

2 Education School, Fuyang Normal University, Fuyang, Anhui 236037, China

Associated Data

The data used to support the findings of this study are available from the corresponding author upon request.

This paper constructs an algorithm for youth school violence recognition and an occupational therapy education model for victims through the extraction of action speech features. For the characteristics of violent actions and daily actions, action features in time and frequency domains are extracted and action categories are recognized by BP neural network; for complex actions, it is proposed to decompose complex actions into basic actions to improve the recognition rate; then, LDA dimensionality reduction algorithm is introduced for the problem of the high complexity of algorithm due to high dimensionality of features, and the feature dimensionality is reduced to 8 dimensions by LDA dimensionality reduction algorithm, which reduces the system running time by about 51% and improves the accuracy of violent action recognition by 3.3% while ensuring the overall performance of the system. The LDA dimensionality reduction algorithm reduces the number of features to 8 dimensions, which reduces the running time of the system by 51%, increases the accuracy rate of violent action recognition by 3.3%, and increases the recall rate of violent action recognition by 8.86% while ensuring the overall performance of the system. Based on the classical D-S theory, we proposed an improved D-S evidence fusion algorithm by modifying the original evidence model with a new probability distribution function and constructing new fusion rules, which can solve the fusion conflict problem well. The recall rate for violent actions is increased to 90.0%, thus reducing the missed alarm rate of the system.

1. Introduction

With increasing media coverage, the phenomenon of school violence has gradually come to the forefront of people's minds. With the development of the Internet, students on campus are exposed to all kinds of information from society, including violence and vulgarity [ 1 ]. Some students, out of curiosity, imitate the behaviors they see on the Internet, making the phenomenon of school violence more serious than ever. Teenagers need proper guidance and education during their formative years. However, school teachers and parents are unable to keep an eye on students, and those who are subjected to violence are often too afraid or shy to report it to teachers and parents promptly [ 2 ]. As a result, violence is not nipped in the bud at the earliest stages of its occurrence, and as a result, it grows worse and worse, seriously affecting the physical and mental health of the victim.

In recent years, there has been a growing trend of school violence in elementary and secondary schools, as reported by the news media. Minor forms of school violence include verbal abuse, pushing, and shoving, while severe forms of school violence include beating and abuse [ 3 ]. A domestic survey showed that, contrary to the perception of a stable school environment, more than 40 percent of students have experienced varying degrees of school bullying. Overseas, a 2010 USA Today survey reported that 50% of high school students surveyed had bullied others in the past year, 47% said they had been bullied, teased, or taunted, and 44% of boys and 50% of girls said they had been victims of school bullying [ 4 ]. This shows that bullying in schools is a very common and serious problem. Violence in schools jeopardizes the normal school and life of the victims and, more seriously, affects their worldview and outlook on life during their formative years. However, the victims of violent bullying are often afraid to give timely feedback to teachers and parents due to fear and self-esteem, thus causing the bullying children to be unable to be reassured and protected on the one hand, and the bullies are not promptly educated and supervised on the other, which eventually makes the phenomenon of school bullying increasingly serious.

A common problem in many school violence cases in recent years is that the bullied students are afraid to inform their parents or teachers about the bullying because they are afraid or threatened, thus causing an indelible psychological shadow on the children's young minds; in addition, the fact that violent incidents are not revealed leads to the perpetrators becoming increasingly rampant, which is also an important factor leading to the increase of school violence [ 5 ]. It is in this context that research on the active detection of violence in schools has emerged. Currently, smartphones are arguably the most widely used wearable sensor devices in the world, which include camera sensors (cameras), GPS sensors, proximity sensors, light sensors, inertial sensors (accelerometers and gyroscopes), and directional sensors (compasses). This topic is based on Video Data Detection of School Bullying aimed at achieving timely and proactive detection and reporting of school bullying by studying the identification of youth school violence based on action-voice features, using devices such as smartphones or bracelets to identify and analyze students' actions and detect the occurrence of violent actions [ 6 ].

2. Related Works

Foreign research on violent action detection was carried out relatively early, and the core problem of violent action recognition is to detect the target human action [ 7 ]. Some early researchers used the relevant features of sound to detect whether there is violence, and some researchers used a combination of video and audio and proposed a violent video detection model based on the semantic information corresponding to the audio and video data in the same video. O'Reilly et al. proposed the first video-based method to identify violent actions by detecting the presence of violence through the detection of blood and flames in images [ 8 ]. Menzies-Gow et al. proposed a violent action detection method using Lagrangian orientation fields to extract features from videos that are based on a spatiotemporal model using appearance, motion compensation of the background, and long-term dynamic information to ensure the scale of spatiotemporal features [ 9 ]. Feldstein et al. chose Gaussian Mixed Model (GMM) technique to extract the candidate regions where violent actions may occur from the information of optical flow and called suspicious activity [ 10 ]. Chen et al. propose to represent human behavior by a set of action units, and at the bottom, they propose a locally weighted word context descriptor to improve the traditional point-of-interest-based representation. This descriptor effectively incorporates neighborhood information [ 11 ]. At the high level, GNMF-based action units are introduced to bridge the semantic gap in action representation. León-Moreno et al. improve system robustness by fusing multisource heterogeneous sensor data, applying information fusion algorithms of fuzzy logic to recognize human behavior in fusion architectures, and performing feature layer fusion to improve recognition rates [ 12 ].

The frequent occurrence of school bullying has led to bad effects making school bullying an issue of widespread concern for the government and society [ 13 ]. Many studies have shown that school bullying is harmful to students for a long time, and as the research on school bullying deepens, school bullying, as a special type of comprehensive problem, is a very different concept from school violence, and domestic scholars have not yet formed a unified concept of school violence. The scholars Zhang et al. refer to school violence as an act in which students are physically and psychologically harmed in some way by teachers, classmates, and people from outside the school [ 14 ]. Li et al.'s view is that all acts of violence against teachers and students that occur within the school's authority are considered school violence [ 15 ]. Nickerson considers school violence to be all violent acts that students are subjected to, whether in public or private schools while attending school, participating in school activities, or after school. Professor Smith of the University of London's view on school bullying is that school bullying is a power struggle, where the strength of the stronger party often can bully and oppress the weaker party, and the bully is usually outnumbered and oppressed repeatedly, with a power mismatch [ 16 ]. Aiming at the problem that the recognition results of different parts of the sensor are seriously conflicting, the optimal fusion is carried out, and an improved D-S theory is proposed for fusion.

The research results on the definition of school bullying and the governance of school bullying at home and abroad are important academic resources for scholars at home and abroad to govern the social problem of school bullying from their different research perspectives, and they are also indispensable theoretical guidance for the practice of school bullying prevention and play an important role in practice [ 17 ]. However, it is easy to find from the literature review that the definition and prevention governance of school bullying emphasizes the legal obligations of the bully and the bullied, psychological construction, and regulations set by the state, while insufficient attention has been paid to the perspective of the motivation of bullying by school bullying subjects to produce school bullying behavior. To address adolescent school violence, this paper focuses on both the perpetrator and the victim and provides occupational therapy education to reduce the likelihood of school violence and its harmfulness.

3.1. Research Design

Construction of an action speech feature-based school violence recognition algorithm and occupational therapy education model for adolescents, whose main research process is to collect human action data by wearing sensors such as accelerometers and gyroscopes on different parts of the body, after which the data are processed and fused and features are extracted to classify and identify school violence actions and give feedback to schools and parents, and the school, family, and society will work together to prevent and stop violence in schools.

The first step is to collect actionable data and preprocess them. Scenario simulation and data for the problem of school violence. This thesis is specifically for this scenario on campus, so as to simulate the campus environment more realistically. Together with the members of the project team, I conducted the experimental data collection for the campus violence scenario. The scenario experiment process included both daily campus actions such as running, jumping, and playing, as well as violent actions such as beating, pushing, and pushing down. Through multiple performances and rehearsals, the campus scenes were realistically restored and motion sensors and voice data were collected.

The second is the study of action feature extraction and feature selection algorithms. Feature extraction is one of the most important processes of pattern recognition and the process that has the greatest impact on recognition rate [ 18 ]. Before extracting the data, preprocessing of the data must be performed. The preprocessing part includes denoising of the received signal and real-time data segmentation. With the same data, different ways of action modeling can correspond to different numbers and types of features, so it is a research challenge to efficiently represent actions with fewer features. The extraction of action features lies in building appropriate action models on the one hand, and on the other hand, filtering and dimensionality reduction of the extracted high-dimensional features is also the key to reducing the complexity of the algorithm. The performance of action recognition based on a single part sensor is obviously inferior to that of multisensor fusion recognition, because multisensors can achieve a more comprehensive capture of actions, and the recognition performance can be improved by fusing different parts. In this paper, the relevant features are extracted in the time domain and frequency domain, and the basic useless features are initially eliminated by using the quadratic box plot method, and then, an improved Relief-F algorithm based on Filter is proposed for feature selection.

The third one is school violence recognition. The BP neural network classifier used in this paper learns through continuous feedback to determine the weights of nodes within different implicit layers and performs several classifier parameters' tuning and training method modifications to complete the classifier design work. For the problem of confusion between violent school actions and daily behavioral actions, a design scheme of joint action-speech feature recognition is proposed.

The fourth is a tripartite school-family-society effort, with schools and families avoiding and reducing the occurrence of school violence and, for those that have occurred, social occupational therapy and educational institutions providing professional psychological treatment for perpetrators and victims to reduce the damaging nature and after-effects of school violence on youth and society. In addition to physical training students, normal adult males can also be used as the subjects of this experiment. To match the theme of youth school violence in the article and fully tap the hidden information of school violence, the second-year physical training students of middle school are the most suitable group.

3.2. Participants

Scenario simulation and data for the campus violence problem. This thesis is specifically aimed at this scenario on campus, so as to simulate the campus environment more realistically. In this paper, through the recruitment of volunteers, 20 physical training students were recruited as participants in this experiment in the second grade of X middle school and randomly combined in groups of 10 people, named group A and group B, respectively.

The rationale for the selection of second-year physical training students from secondary school X in this paper is as follows.

• Both age and physical characteristics were perfectly matched to the adolescent school violence perpetrators and victims
• Physical training students are solid in all aspects of physical fitness due to years of sports training. They can very well complete the daily running, jumping, playing, and other campus action at the same time, complete the beating, pushing, and pushing down, and another violent action is not easy to be injured, and can ensure the maximum physical safety of the experimental participants
• The second year of middle school is a critical time in the formation of life and values, and this age group is most likely to be involved in school violence

3.3. Measures

3.3.1. data accounting aspects.

The number of network layers and the number of neurons were set for the neural network using the filtered and dimensionality-reduced features as the input to the neural network [ 19 ]. The transfer functions used in the implicit and output layers were determined after several trials. After that, we added the “fall” and “push” actions and reselected the features to determine the relevant parameters of the neural network. To address the problem that “hitting” actions are easily confused with nonviolent actions, the “hitting” actions were decomposed, and the data collection experiments of the decomposed actions were carried out to analyze the data and make the classification.

3.3.2. Action Speech Recognition Aspects

Speech and action features are combined to further improve the action recognition rate. Firstly, the theory related to MFCC (Mel Frequency Cepstrum Coefficient) coefficients, which are very important in speech features, is introduced, and the MFCC and short-time energy features are extracted and classified for the speech data collected in the action test, and the results show that the classification effect of speech-based violence is much lower than that of motion sensor-based classification. Out of curiosity, some students imitated behaviors seen on the Internet, making the phenomenon of school violence more serious than ever.

3.3.3. System Optimization Aspects

To optimize the built system, firstly, for the high time complexity of the system due to the excessive number of selected feature dimensions, the system is optimized by combining the dimensionality reduction algorithm to reduce the number of feature dimensions, which significantly reduces the recognition time of the system and lays the foundation for future hardware implementation. In addition, the problem of serious conflicts between the recognition results of different parts of the sensor is optimized fusion and proposed to improve the D-S theory for fusion, simulation analysis, and adaptive adjustment of the decision layer class fusion algorithm for comparison and analysis of the conclusions.

3.4. Design

The study is based on multisensor fusion recognition, so the location and number of sensors are crucial. The study shows that the built-in sensors of wearable devices are located in different locations, and the number of sensors will have a direct impact on the recognition results. For different actions, different locations of sensors will have different recognition accuracy [ 20 ]. At present, most of the recognition of human actions are located at the waist, legs, wrists, and chest; in addition, for the fusion recognition system of multiple sensors, there are still multiple choices of combinations of different locations. “Lying,” “sitting,” “standing,” “walking,” “running,” “going upstairs,” “going downstairs,” “jogging,” “jumping,” and other daily movements, the current common positions, and combinations of study results in specific performance comparisons are shown in Table 1 . Through comparison, it can be found that the performance of action recognition based on a single part sensor is inferior to multisensor fusion recognition, which is because multisensor can achieve more comprehensive capture of action, and fusing different parts can improve recognition performance. In addition, the combination of different parts has a greater impact on recognition accuracy, and research shows that the more sensors are not the higher the recognition accuracy, and the number of sensors worn needs to consider the balance of convenience and recognition rate. In this project, the data was collected by wearing sensors on the waist, legs, and wrists, and combined with the current research, it was decided to use the sensors on the waist and legs for fusion detection of violent campus movements. Therefore, in this study, the motion sensor was placed on the waist of the experimenter, and the triaxial acceleration and triaxial gyroscope signals were collected.

Performance comparison of multisensor position combinations.

The action recognition system mainly consists of the following parts: data acquisition and processing, feature extraction and selection, and classifier design. At present, the common idea of data segmentation is to set a sliding window, and the data collected by the sensor is stored in the form of the data stream, by choosing a suitable window length, sliding in the time axis of the data stream according to a certain ratio and speed can be. In this paper, the window length is set to 256 sampling points, and the sampling frequency of the collected data sensor is set to 50 Hz, so the time length of each extraction is about 5 s. After comparison, the data within this time length can fully reflect the change of action and can meet the requirements of final recognition. In addition, since the feature extraction needs to be analyzed in the frequency domain afterward, the design length of the sliding window is chosen as a power of 2 for the convenience of time-frequency conversion.

Victims of violence often do not dare to report the situation to teachers and parents in a timely manner out of fear or shyness. This leads to the fact that in the early stages of violence, violence cannot be nipped in the bud. BP neural networks are no longer suitable for places with high real-time requirements due to their global approximation nature and thus very slow learning speed, and the deep learning-based Convolution Neural Network (CNN) has the advantages of no manual feature selection and excellent classification effect, but the training network requires a large amount of sample data, which is very demanding for equipment performance requirements are very high. Compared with BP and Convolutional Neural Network, Radial Basis Function (RBF) has the advantages of simple network structure, fast convergence, and the ability to fit any nonlinear function, so it is widely used in various fields [ 21 ].

Since the acquisition of action data is affected by many noises, for this reason, it is necessary to choose a suitable method for smoothing and denoising. Several commonly used filtering and denoising methods are mean, median, smoothing, Gaussian, and low-pass. Among them, smoothing filtering is easy to operate and simple to understand. In a segment of data, the average of the neighboring points is sought, and the size of its selected neighborhood cannot be too large; otherwise, it will produce information loss. Median filtering is to select the middle value to avoid the influence of noise, which will not produce edge information loss, simple calculation, and is easy to implement with hardware. Mean filtering replaces the average value of multiple points around a point with a small number of points that will have large fluctuations in the average value. Gaussian filtering is replaced with a weighted average of its own and the data in the field, yielding a broken flat edge problem. Low-pass filtering can be thought of as setting a fixed frequency above which the frequency domain is filtered out and vice versa, which is allowed to pass. Since the human body movement frequency is relatively small, the normal movement is between 1 and 50 Hz. Therefore, the IIR Butterworth (Butterworth) low-pass filter is selected to remove the noise, calculated as follows.

To make the processing of the collected data easier, the data is normalized. The data is scaled to make the data in a certain limited area. Normalization is a dimensionless processing method that makes the values relative and units the data with different ranges so that they have a uniform distribution and are easier to calculate and understand. The minimum-maximum normalization method of normalization, also known as discrete normalization, is used in this paper.

Since each person completes an action with a different amplitude and frequency, the length of the data collected by the sensor is not the same, and it takes a relatively long time to collect action and accumulate a large amount of data. For this reason, the collected data need to be regularized and segmented into different windows, with adjacent windows partially overlapping, and the features of the windows are extracted to provide an accurate description of the human action characteristics. Three common window addition methods are sliding window segmentation, event-based definition window segmentation, and action-based window segmentation.

The time length of each extraction is about 5 s. After comparison, the data within this time length can fully reflect the change of an action, which can meet the final identification requirements. For the input of the deep learning network, this section adopts the strategy of dividing the original audio waveform into frames and then using the corresponding waveform map of each frame as the input of the AlexNet network to extract the audio features of the original waveform. Finally, the audio features are fed into the LSTM network for modeling the timing signal. The output of the last LSTM unit is the memory of the most effective features of the whole audio segment, after which the FC layer is connected to classify the audio segment effectively. Block diagram of the violence audio detection system based on the original audio waveform is shown in Figure 1 .

Block diagram of the violence audio detection system based on the original audio waveform.

3.5. Analysis

The DS fusion algorithm, when faced with severely conflicting evidence, fuses the result as a violent event and ignores the credibility of voice evidence; the Yager fusion algorithm, when faced with severely conflicting evidence, synthesizes the result showing that the Yager fusion process reduces the credibility of both sets of evidence, and while assigning more support to the uncertainty function, i.e., it does not make an exact judgment, which is not applicable in practical scenarios. Compared with the improved Yager synthesis result, the proposed improved solution reduces the credibility of video evidence and increases the credibility of voice evidence, but the synthesis rule in this paper assigns conflicts to the focal elements that generate conflicts, while the improved Yager assigns conflicts to all focal elements. The improved Yager fusion algorithm is more reasonable than the improved Yager fusion algorithm. On the one hand, the bullied children cannot be comforted and protected, and on the other hand, the bullies have not received timely education and supervision, which eventually makes the phenomenon of school bullying increasingly serious.

In the network training process, we only need to train the parameters of the LSTM network since the parameters of the CNN (AlexNet) network that takes the features are trained. The maximum number of iterations is set to 100, and the loss value varies with the number of iterations. To characterize the envelope of the audio in the long-time range, we also take the statistical features. The violent audio detection method based on acoustic features and long-time statistical features achieves better detection results. In terms of deep learning, this paper adopts two strategies, original waveform as network input and audio speech spectrogram as network input, and after comparison, we find that the end-to-end detection method based on the original audio waveform is more beneficial to violent audio detection. This also provides a new idea for the later audio-video feature fusion method.

The background of this study is for school violence scenarios, to truly simulate school violence actions to ensure the practicality of the recognition system, for complex violent actions through the protection measures to ensure that the data is real and reliable, and the use of several different experimental subjects to repeatedly simulate various types of actions, so that the collection database is more convincing. The collected actions include “hitting,” “pushing,” “pushing down,” and other common school violence actions and “running,” “playing,” and “jumping.” The first step is to identify these nine types of actions. These nine types of actions are identified, and finally, the performance indexes are measured according to violent and nonviolent actions. Not all the proposed features are useful for classification and recognition, after which they are firstly screened by using quadratic box plots to distinguish useless features for actions and then studied by using feature selection algorithms. A comparative study was conducted on two groups of physical training students, A and B. The test index data of the experimental group were analyzed by mathematical and statistical methods. The experimental results demonstrate that the fusion algorithm has improved the recognition rate of actions compared to any single part sensor, and the recognition rate is 85.3%, especially for violent actions after the fusion. The recognition rate of complex actions such as “fall,” “push,” “push,” and “hit” is improved by 7.2%. By calculating the four index parameters, we can see that the overall system performance is improved by about 4.98%. The solution will also be proposed for the fusion problem in the case of conflicting sensor recognition of different parts, to improve the recognition rate of complex actions such as “fall” and “push down.” In this paper, “push” and “fall” actions are added. However, the action process of fall is short, as shown in Figure 2 , which shows the change of acceleration sensor during the occurrence of violent action.

Schematic diagram of acceleration sensor changes for fall-like movements.

4.1. Motion Speech Feature Extraction

A 5-fold cross-validation was used to test the performance of the model, and the test samples were divided into 5 equal parts, and the average of the 5 recognition results was taken as the result, and four metrics, accuracy, precision, recall, and F1-score, were selected to evaluate the classification algorithm. The test and evaluation results for the three audio databases are shown in Table 2 .

Speech emotion classification test results.

Table 2 shows that the classification algorithm achieves a correct rate of 88.33% on the self-made speech database, the highest correct recognition rate of 95% on the Finnish speech database, and 91.67% on the CASIA public speech database, which indicates that the classification algorithm has a better performance in recognizing speech emotion. In addition, the recall rate on all three speech databases is above 85%, which indicates that the F1-score of the classification algorithm is the highest for the Finnish speech database, followed by the CASIA database, which indicates that the classification algorithm has the best performance on the Finnish speech database, followed by the CASIA database and the homemade speech database. The reason for the analysis is that for the home-made small speech database, the subject members had poor performance in expressing emotions through speech, which resulted in the difference between bullying emotions and nonbullying emotions not being obvious, which had an impact on the final correct recognition rate. However, the Finnish speech database is a school bullying and nonbullying scenario simulated by elementary school students, which is very suitable for the needs of this topic, so the classification algorithm performs best on this database. The CASIA speech database is recorded by professionals, and the emotions expressed by speech are fuller, and the differences in emotions expressed by different speech are larger, so the differences in feature vectors are also larger. The classification algorithm performs better on the Finnish speech database and the CASIA speech database.

In the experiment, 14 kinds of action data were collected for each person, and there were 20 people in total. 15 people's action data from each action were selected as training set samples, and 5 people's action data were selected as test set samples, and there were 210 samples in the training set and 70 samples in the prediction set. In the previous mathematical theoretical derivation, it has been found that the penalty factor and radial kernel function parameters are two important parameters for SVM classification performance. Therefore, this paper discusses the comparison of SVM classification performance after four random choices of these two parameters and after PSO parameter optimization ( Figure 3 ). Radial Basis Function (RBF) network has the advantages of simple network structure, fast convergence speed, and the ability to fit arbitrary nonlinear functions, so it has been widely used in various fields. Comparison of SVM classification performance after PSO parameter optimization: the results of four times random parameter search under time-domain features have been compared with the SVM classification performance after the PSO parameter optimization process, and the range of random selection process c and g is (0,100). The random selection of different two important parameters directly affects the classification performance of the SVM. As the number of iterations increases, the red best-fit curve shows an increasing trend of classification recognition rate, which reaches the optimum and remains unchanged when it reaches 8 iterations, with an accuracy of 96.97% under fivefold crossover, a penalty factor c of 1.6294, and a radial kernel function g of 11.0116, and the SVM classification model is built with these two optimal parameters, which finally obtains a classification recognition rate of 92.85%.

PSO-SVM parameter search process in the time domain.

4.2. Campus Violence Identification

In the conversion rate of violent action recognition, because the amplitude of motion is close to the interaction between violent action and running motion, to improve its action recognition accuracy, a method of joint recognition by multilayer classification algorithm should be designed, and then, the extracted features are analyzed and processed. Firstly, the SVM algorithm is used to recognize them, and then, the threshold decision tree classification algorithm is used in the decision layer to further classify the classification level. Finally, the two prediction results are combined and then weighted, the weights of the features are reasonably assigned, and finally, a DT-SVM multilayer classification model is established to improve the action recognition rate. Here, the selected features are first analyzed to find out the decision threshold of a certain type of action. Since some features with certain thresholds can directly distinguish violent and nonviolent behaviors, for example, the spacing of the center of mass and the magnitude of the action, the probability of nonviolent behaviors can be directly determined when there is no obvious action in the video and the active participants are very far away, or the probability of violent behaviors is significantly increased when one of the action participants falls. In this paper, relevant features are extracted in the time domain and frequency domain, respectively, and the basic useless features are initially eliminated by the quartile box plot method, and then, an improved Relief-F algorithm based on Filter is proposed for feature selection. On the premise of the recognition rate, the feature dimension is reduced by the dimensionality reduction algorithm for system optimization.

A total of 12448 violent action frames, 9963 nonviolent action video frames, and 2485 nonviolent undetected target video frames were collected. A total of 17 features were extracted, and the feature selection algorithm was used to select the optimal parameters after filtering the features and thresholds for recognition using fivefold cross-validation. The results show that 96.52% of violent actions are accurately identified as violent actions, and 3.48% are incorrectly identified as nonviolent actions. 80.81% of nonviolent actions are accurately identified as nonviolent actions, and 19.19% of nonviolent actions are incorrectly identified as violent actions. The final obtained classification accuracy using only the SVM classification algorithm was 89.54%, accuracy was 94.90%, the recall was 80.81%, and algorithm performance parameter F1_S was 87.29%. As shown in Figure 4 , compared to the DT-SVM algorithm with only the inclusion of video frames with obvious actions, the DT-SVM algorithm with the inclusion of some undetected target frames has 0.78% higher accuracy, 2.14% higher precision, 0.74% higher recall, and 1.31% higher algorithm performance. The undetected target frames can be detected well in the foreground detection stage, and the partial undetected target frames are added only to demonstrate the algorithm's classification of undetected target frames.

Quartile plot of y -axis variance characteristics of waist data.

4.3. Constructs of Occupational Therapy Education Models

Studies have found that the development of adolescent personality is closely related to family education and living environment. Children who lack parental companionship tend to be withdrawn and quiet. Especially children left behind, because their parents are not around, food, clothing, housing, and transportation are not well taken care of, and even when they are bullied, they have no place to cry because their parents are not around, and overall, they develop a withdrawn, introverted, and timid personality. In addition, without parents to “back up,” they are bullied and have no way to complain and gradually become victims of school violence.

In the rapidly developing economic era, most parents are only concerned with making money to support their families and believe that their children only need to study hard and have stable and excellent academic performance. Parents rarely participate in their children's growth process and lack the necessary communication with their children. Children increasingly reject open communication with their parents or fear communication with their parents, which may lead to certain psychological disorders. At the same time, parents' bad habits may also set the wrong values for their children, for example, parents often drink in front of their children and quarrel and fight with each other, and children will learn from their parents' bad habits and become violent over time. What is more, some parents lack of patience, and the child made a mistake, indiscriminately is a beating, and eventually, the child encountered anything also choose to save violence, and such children gradually become the perpetrators of school violence. To distinguish between school violence and daily actions, this paper designs an action-speech joint feature recognition model. The model is mainly designed to be equipped with sound-collecting sensors. Violence in schools is often accompanied by the bully's loud abuse and the victim's crying. And the large-scale movements such as running, jumping, and falling are not accompanied by insults and crying.

Middle school students are in early adolescence. Most adolescents are psychologically adventurous and impulsive, and they like to imitate others' words, actions, and dresses. Most adolescents are adventurous and impulsive, and they like to imitate others' words, behaviors, and dresses. This period of adolescence is accompanied by many shortcomings, such as more contact with social things but lack of social experience, relatively reckless and lack of rational thinking and even reckless to achieve some immature purposes, weak ability to distinguish right from wrong, and easily influenced by the negative environment around them and the violent factors in movies and online games. If we do not control and guide them correctly, they will easily go astray and do things that are detrimental to their good development and even endanger social security in serious cases.

Different from previous studies that focused unilaterally on victims of school violence, this paper analyzes the psychological factors of both perpetrators and victims and constructs a perpetrator-victim occupational therapy education model for the different psychologies of perpetrators and victims, as shown in Figure 5 .

Perpetrator-victim occupational therapy education model.

In response to incidents of school violence that have occurred, schools, families, and society should act proactively and cooperate to provide reasonable psychological guidance to perpetrators and victims of violence to avoid lifelong psychological trauma.

For the abuser . Abusers themselves often have deficiencies in emotional communication and self-behavior control, such as a lack of proper emotional communication skills, empathy, and proper self-behavior control methods, leading them to find an outlet to vent and express themselves and eventually to take the path of bullying others. The important reason we need to pay attention to and take effective measures to help the bully improve their own words and behaviors is based on understanding and respecting each individual, guiding them to establish a correct sense of communication and interaction based on understanding the motivation of the bully's behavior, and providing some methods and skills to improve their own words and behaviors, so that the bully knows the harm of bullying others and that the bullying behavior can be corrected, and so that the bully will feel accepted, understood, and trusted and will then take the initiative to improve his or her behavior and speech.

Targeting victims . A generalized analysis of students' responses to violence on secondary school campuses shows that more than half of the victims chose to keep quiet after being subjected to violence on school campuses, and they chose to keep quiet because they were afraid of the abuser's reoccurrence. Very few of the victims were affected by the violence, but their normal school life was affected to a certain extent. Some students who are victims of campus violence have mental problems, such as fear and suspicion all day long, serious fear of the campus and the abuser, and are afraid to come to school. Some students even have a nervous breakdown and become mentally ill. Some students choose to end their lives to escape the violence. For victims of school violence, it is important to focus on psychological and emotional relief in addition to physical treatment. It is important to guide the victims to come out of the gloom, believe in society again, and work hard to live a positive life. It is also important to instill a sense of courage to speak out in the face of school violence. If violence occurs on campus, report it to the teacher first to receive appropriate assistance, and do not be reckless and retaliate against the perpetrator to vent your anger and prevent the violence from escalating again. It is best to avoid being alone on the way to and from school and to avoid choosing roads with few people and remote locations. When you go out alone, pay attention to strangers, do not meet with people you know online, and do not go to Internet cafes, game halls, and other entertainment places that you are not allowed to enter. In short, only by learning to protect yourself and having a strong sense of protection can you avoid being harmed by school violence.

5. Discussion

Usually, when the topic of school bullying is mentioned, people will express their opposition or hatred, putting the bully in the position of the weak and the bully in the position of the strong. However, both the bully and the bullied in school bullying incidents are in a vulnerable position, and neither of them uses the right ways or methods to express their emotions and needs and maintain their interpersonal relationships. The bullies themselves often have deficiencies in emotional communication and self-behavior control, such as a lack of proper emotional communication skills, empathy, and proper self-behavior control methods, resulting in their need to find an outlet to vent and express themselves and eventually taking the path of bullying others. The important reason why we should pay attention to and take effective measures to help bullies improve their own words and behaviors is based on understanding and respecting each individual, guiding them to establish a correct sense of communication and interaction based on understanding the motivation of the bully's behavior, and providing some methods and techniques to improve their own words and behaviors, so that the bully knows the harm of bullying others and that the bullying behavior can be corrected, so that the bully will feel accepted, understood, and trusted and will then actively participate in improving their speech and behavior, experience the positive effects of improving their speech and behavior, and then eliminate the bullying behavior.

6. Conclusion

In this paper, we studied the action characteristics of violent actions and collected data on “push,” “push down,” “squat,” and “roll.” The simulation shows that the average recognition accuracy is 88.35%, and the recall rate of violent action recognition is 84.63%. To address the feature redundancy problem of the Relief-F feature selection algorithm, a redundancy improvement algorithm is proposed, a multilayer classifier based on decision tree-RBF neural network is built, and a PACBF algorithm is designed for adaptive fusion processing in the decision layer. For the problem of fusion conflict of multipart recognition results, an improved D-S theoretical fusion model is proposed to modify the original evidence model by designing a new distribution function and constructing a new fusion rule, which finally achieves a system recognition rate of 93% and a recall rate of 91%, greatly reducing the system leakage rate. The occupational therapy model is theoretically able to address the negative effects of school violence on both perpetrators and victims, but due to time limitations, the selection of psychology professionals is not ideal, and the next work can be conducted in more detail and depth in this area.

Acknowledgments

This work was supported by the Institute of Education, Joongbu University.

Data Availability

Conflicts of interest.

The authors declare that there is no conflict of interest.

Missouri teen beaten in viral video is out of ICU but has limited speech and trouble walking on her own, attorney says

A Missouri teenager who was brutally beaten in what officials called a "deranged display of violence" by another teen is out of the intensive care unit but has limited speech and trouble walking on her own, an attorney for the family said.

Kaylee Gain has been hospitalized since a March 8 fight near Hazelwood East High School in St. Louis County that was captured in a viral social media video.

The footage shows several people brawling in the street near the intersection of Norgate and Claudine drives, the St. Louis County Police Department said in a  March 11 Facebook post .

One person is seen repeatedly punching Gain and slamming her head to the ground. A 15-year-old girl was arrested on assault charges a day after the fight, authorities said.

Police said the victim was found "suffering a severe head injury" and was taken to the hospital in critical condition.

In an update Friday, an attorney for Gain's family said she was out of the intensive care unit and "has been able to engage in limited verbal conversations."

"Kaylee also recently began speech therapy, and has gone on a few short walks with the assistance of hospital staff as she is still unable to ambulate on her own," attorney Bryan Kaemmerer said. "However, Kaylee does not have any recollection of the altercation that led to her hospitalization."

Kaemmerer addressed several social media rumors about the altercation, denying reports that Gain's mother drove her to the location of the fight.

He said Gain's mother was at work and was driven to the hospital by a co-worker after police informed her of what happened.

The attorney, however, did confirm reports that Gain had been involved in a fight on March 7 with a different teenager. Both girls were suspended after that incident, Kaemmerer said.

He said it was unclear whether the March 8 brawl was retaliation.

Gain's parents are calling for the 15-year-old to be tried as an adult. Kaemmerer said in his statement that "the family believes trying the accused as an adult is the most appropriate way to provide the justice that Kaylee deserves."

Authorities have not said if the 15-year-old would be tried as an adult.

St. Louis County Prosecuting Attorney Wesley Bell  said  in a post on X that the fight was "sickening" and the video was "difficult to watch."

Missouri Attorney General Andrew Bailey called the actions in the video a "deranged display of violence that must be punished to the full extent of the law."

On Thursday, police announced that eight more teenagers were referred to St. Louis County Family Court for consideration of assault charges, NBC affiliate KSDK of St. Louis reported. They include a 17-year-old girl, a 17-year-old boy, two 16-year-old girls, three 16-year-old boys, and one 14-year-old girl. None of the teens have been taken into custody.

Minyvonne Burke is a senior breaking news reporter for NBC News.