Is hate speech detection the solution the world wants?

The machine learning (ML) research community has landed on automated hate speech detection as the vital tool in the mitigation of bad behavior online. However, it is not clear that this is a widely supported view outside of the ML world. Such a disconnect can have implications for whether automated detection tools are accepted or adopted. Here we lend insight into how other key stakeholders understand the challenge of addressing hate speech and the role automated detection plays in solving it. To do so, we develop and apply a structured approach to dissecting the discourses used by online platform companies, governments, and not-for-profit organizations when discussing hate speech. We find that, where hate speech mitigation is concerned, there is a profound disconnect between the computer science research community and other stakeholder groups-which puts progress on this important problem at serious risk. We identify urgent steps that need to be taken to incorporate computational researchers into a single, coherent, multistakeholder community that is working towards civil discourse online.

On the rise of fear speech in online social media.

Recently, social media platforms are heavily moderated to prevent the spread of online hate speech, which is usually fertile in toxic words and is directed toward an individual or a community. Owing to such heavy moderation, newer and more subtle techniques are being deployed. One of the most striking among these is fear speech. Fear speech, as the name suggests, attempts to incite fear about a target community. Although subtle, it might be highly effective, often pushing communities toward a physical conflict. Therefore, understanding their prevalence in social media is of paramount importance. This article presents a large-scale study to understand the prevalence of 400K fear speech and over 700K hate speech posts collected from Gab.com. Remarkably, users posting a large number of fear speech accrue more followers and occupy more central positions in social networks than users posting a large number of hate speech. They can also reach out to benign users more effectively than hate speech users through replies, reposts, and mentions. This connects to the fact that, unlike hate speech, fear speech has almost zero toxic content, making it look plausible. Moreover, while fear speech topics mostly portray a community as a perpetrator using a (fake) chain of argumentation, hate speech topics hurl direct multitarget insults, thus pointing to why general users could be more gullible to fear speech. Our findings transcend even to other platforms (Twitter and Facebook) and thus necessitate using sophisticated moderation policies and mass awareness to combat fear speech.

Disrupting hate: The effect of deplatforming hate organizations on their online audience.

How does removing the leadership of online hate organizations from online platforms change behavior in their target audience? We study the effects of six network disruptions of designated and banned hate-based organizations on Facebook, in which known members of the organizations were removed from the platform, by examining the online engagements of the audience of the organization. Using a differences-in-differences approach, we show that on average the network disruptions reduced the consumption and production of hateful content, along with engagement within the network among periphery members. Members of the audience closest to the core members exhibit signs of backlash in the short term, but reduce their engagement within the network and with hateful content over time. The results suggest that strategies of targeted removals, such as leadership removal and network degradation efforts, can reduce the ability of hate organizations to successfully operate online.

Empathy-based counterspeech can reduce racist hate speech in a social media field experiment.

Despite heightened awareness of the detrimental impact of hate speech on social media platforms on affected communities and public discourse, there is little consensus on approaches to mitigate it. While content moderation-either by governments or social media companies-can curb online hostility, such policies may suppress valuable as well as illicit speech and might disperse rather than reduce hate speech. As an alternative strategy, an increasing number of international and nongovernmental organizations (I/NGOs) are employing counterspeech to confront and reduce online hate speech. Despite their growing popularity, there is scant experimental evidence on the effectiveness and design of counterspeech strategies (in the public domain). Modeling our interventions on current I/NGO practice, we randomly assign English-speaking Twitter users who have sent messages containing xenophobic (or racist) hate speech to one of three counterspeech strategies-empathy, warning of consequences, and humor-or a control group. Our intention-to-treat analysis of 1,350 Twitter users shows that empathy-based counterspeech messages can increase the retrospective deletion of xenophobic hate speech by 0.2 SD and reduce the prospective creation of xenophobic hate speech over a 4-wk follow-up period by 0.1 SD. We find, however, no consistent effects for strategies using humor or warning of consequences. Together, these results advance our understanding of the central role of empathy in reducing exclusionary behavior and inform the design of future counterspeech interventions.

Desensitization to hate speech: Examination using heart rate measurement.

Exposure to hate speech (HS) leads to desensitization of listeners. Yet, most evidence of this process has been obtained using self-report measures. In this paper, we examined desensitization to HS using an unobtrusive, psychophysiological measure. In an experimental electrocardiogram study (N = 56), we observed heart rate (HR) deceleration after reading comments that contained HS. This suggested a substantive psychophysiological reaction of participants to hateful comments. However, such HR deceleration was not observed among participants preexposed to HS. People exposed to hateful comments thus appeared to show different HR responses to HS compared to people who were not previously exposed to such comments. Consequently, not only does frequent exposure to HS influence an individual's beliefs as observed in earlier studies, but it also impacts psychophysiological reactions to derogatory language.

Validation of the multidimensional bystander responses to racist hate speech scale and its association with empathy and moral disengagement among adolescents.

Our understanding of how bystanders respond to hate speech is limited. This may be due, in part, to the lack of available measurement tools. However, understanding adolescents' responses to hate speech is critical because this kind of research can support schools in empowering students to exhibit courageous moral behavior. Thus, the purpose of the present study was to investigate the psychometric properties of the newly developed Multidimensional Bystander Responses to Hate Speech Scale (MBRHS) and to explore demographic differences and correlates of bystander behavior in school hate speech. The sample consisted of 3225 seventh to ninth graders from Germany and Switzerland. Exploratory and confirmatory factor analyses supported a model with seven factors. We found that adolescents with immigrant background and boys showed particularly unfavorable response patterns. In addition, our study suggests that empathy is positively correlated with the factors comforting the victim, seeking help at school, and countering hate speech but negatively correlated with helplessness, revenge, reinforcing, and ignoring. Moral disengagement showed the opposite correlational pattern. The findings indicate that the MBRHS is a psychometrically valid and reliable measure that could aid in measuring varied responses to hate speech. In addition, this work highlights the relevance of empathy and moral engagement training in anti-hate speech prevention programs.

Direct and indirect effects of social dominance orientation on hate speech perpetration via empathy and moral disengagement among adolescents: A multilevel mediation model.

Although it is known that social dominance orientation directly affects hate speech perpetration, few studies have explored the mechanisms by which this effect takes place during adolescence. Based on the socio-cognitive theory of moral agency, we aimed to fill this gap in the literature by exploring the direct and indirect effects of social dominance orientation on hate speech perpetration in offline and online settings. The sample included seventh, eigth, and ninth graders (N = 3225) (51.2% girls, 37.2% with an immigrant background) from 36 Swiss and German schools who completed a survey about hate speech, social dominance orientation, empathy, and moral disengagement. A multilevel mediation path model revealed that social dominance orientation had a direct effect on offline and online hate speech perpetration. Moreover, social dominance also had indirect effects via low levels of empathy and high levels of moral disengagement. No gender differences were observed. Our findings are discussed regarding the potential contribution to preventing hate speech during adolescence.

Enhancing Monitoring Performance: A Microservices Approach to Monitoring with Spyware Techniques and Prediction Models.

In today's digital landscape, organizations face significant challenges, including sensitive data leaks and the proliferation of hate speech, both of which can lead to severe consequences such as financial losses, reputational damage, and psychological impacts on employees. This work considers a comprehensive solution using a microservices architecture to monitor computer usage within organizations effectively. The approach incorporates spyware techniques to capture data from employee computers and a web application for alert management. The system detects data leaks, suspicious behaviors, and hate speech through efficient data capture and predictive modeling. Therefore, this paper presents a comparative performance analysis between Spring Boot and Quarkus, focusing on objective metrics and quantitative statistics. By utilizing recognized tools and benchmarks in the computer science community, the study provides an in-depth understanding of the performance differences between these two platforms. The implementation of Quarkus over Spring Boot demonstrated substantial improvements: memory usage was reduced by up to 80% and CPU usage by 95%, and system uptime decreased by 119%. This solution offers a robust framework for enhancing organizational security and mitigating potential threats through proactive monitoring and predictive analysis while also guiding developers and software architects in making informed technological choices.

Individual and Contextual Correlates of Latent Bystander Profiles toward Racist Hate Speech: A Multilevel Person-centered Approach.

Prior research into bystander responses to hate speech has utilized variable-centered analyses - such approaches risk simplifying the complex nature of bystander behaviors. Hence, the present study used a person-centered analysis to investigate latent hate speech bystander profiles. In addition, individual and classroom-level correlates associated with the various profiles were studied. The sample included 3225 students in grades 7-9 (51.7% self-identified as female; 37.2% with immigrant background) from 215 classrooms in Germany and Switzerland. The latent profile analysis revealed that four distinct profiles could be distinguished: Passive Bystanders (34.2%), Defenders (47.3%), Revengers (9.8%), and Contributors (8.6%). Multilevel logistic regression models showed common and distinct correlates. For example, students who believed that certain social groups are superior were more likely to be Revengers and Contributors than Passive Bystanders, students who felt more connected with teachers were more likely to be Defenders, and students who were more open to diversity were less likely to be Contributors than Passive Bystanders. Students were less likely Defenders and more likely Revengers and Contributors than Passive Bystanders in classrooms with high rates of hate speech perpetration. Further, in classrooms with high hate speech intervention, students were more likely to be Defenders and less likely to be Contributors than Passive Bystanders. In classrooms with stronger cohesion, students were more likely to be Defenders and less likely to be Contributors than Passive Bystanders. In conclusion, the findings add to our understanding of bystander profiles concerning racist hate speech and the relevance of individual and classroom-level factors in explaining various profiles of bystander behavior.

Hate Speech Against Asian American Youth: Pre-Pandemic Trends and The Role of School Factors.

Although hate speech against Asian American youth has intensified in recent years-fueled, in part, by anti-Asian rhetoric associated with the COVID-19 pandemic-the phenomenon remains largely understudied at scale and in relation to the role of schools prior to the pandemic. This study describes the prevalence of hate speech against Asian American adolescents in the US between 2015 and 2019 and investigates how school-related factors are associated with whether Asian American youth are victims of hate speech at school. Analyses are based on a sample of 938 Asian American adolescents (Mage = 14.8; 48% female) from the three most recently available waves (2015, 2017, and 2019) of the School Crime Supplement to the National Crime Victimization Survey. On average, approximately 7% of Asian Americans were targets of hate speech at school between 2015 and 2019, with rates remaining stable over time. Findings also indicate that students had lower odds of experiencing hate speech if they attended schools with a stronger authoritative school climate, which is characterized by strict, yet fair disciplinary rules coupled with high levels of support from adults. On the other hand, Asian American youth faced higher odds of experiencing hate speech if they were involved in school fights. Authoritative school climate and exposure to fights are malleable and can be shaped directly by broader school climate related policies, programs and interventions. Accordingly, efforts to promote stronger authoritative climates and reduce exposure to physical fights hold considerable potential in protecting Asian American youth from hate speech at school.

Multilingual Hate Speech Detection: A Semi-Supervised Generative Adversarial Approach.

Social media platforms have surpassed cultural and linguistic boundaries, thus enabling online communication worldwide. However, the expanded use of various languages has intensified the challenge of online detection of hate speech content. Despite the release of multiple Natural Language Processing (NLP) solutions implementing cutting-edge machine learning techniques, the scarcity of data, especially labeled data, remains a considerable obstacle, which further requires the use of semisupervised approaches along with Generative Artificial Intelligence (Generative AI) techniques. This paper introduces an innovative approach, a multilingual semisupervised model combining Generative Adversarial Networks (GANs) and Pretrained Language Models (PLMs), more precisely mBERT and XLM-RoBERTa. Our approach proves its effectiveness in the detection of hate speech and offensive language in Indo-European languages (in English, German, and Hindi) when employing only 20% annotated data from the HASOC2019 dataset, thereby presenting significantly high performances in each of multilingual, zero-shot crosslingual, and monolingual training scenarios. Our study provides a robust mBERT-based semisupervised GAN model (SS-GAN-mBERT) that outperformed the XLM-RoBERTa-based model (SS-GAN-XLM) and reached an average F1 score boost of 9.23% and an accuracy increase of 5.75% over the baseline semisupervised mBERT model.

Label modification and bootstrapping for zero-shot cross-lingual hate speech detection.

The goal of hate speech detection is to filter negative online content aiming at certain groups of people. Due to the easy accessibility and multilinguality of social media platforms, it is crucial to protect everyone which requires building hate speech detection systems for a wide range of languages. However, the available labeled hate speech datasets are limited, making it difficult to build systems for many languages. In this paper we focus on cross-lingual transfer learning to support hate speech detection in low-resource languages, while highlighting label issues across application scenarios, such as inconsistent label sets of corpora or differing hate speech definitions, which hinder the application of such methods. We leverage cross-lingual word embeddings to train our neural network systems on the source language and apply them to the target language, which lacks labeled examples, and show that good performance can be achieved. We then incorporate unlabeled target language data for further model improvements by bootstrapping labels using an ensemble of different model architectures. Furthermore, we investigate the issue of label imbalance in hate speech datasets, since the high ratio of non-hate examples compared to hate examples often leads to low model performance. We test simple data undersampling and oversampling techniques and show their effectiveness.

The relation of classroom climate to adolescents' countering hate speech via social skills: A positive youth development perspective.

INTRODUCTION: Hate speech is a current challenge for schools around the globe. At the same time, students worldwide stand up to hate speech by countering it. Guided by a positive youth development perspective, the present study investigated the direct and indirect associations between classroom climate (environmental assets), social skills (personal assets), and countering hate speech (as a proxy of thriving) among adolescents. METHODS: The sample included 3225 students in grades 7-9 (51.7% self-identified as female) from 40 schools in Germany (n = 1841) and Switzerland (n = 1384). Students completed self-report questionnaires that assessed classroom climate, three facets of social skills (i.e., perspective-taking, prosocial behavior, assertiveness), and counterspeech. RESULTS: The results of the 2-(1-1-1)-1 multilevel mediation analysis revealed that classroom climate (L2) and the three facets of social skills (L1) had a direct positive effect on counterspeech (L1). Furthermore, classroom climate (L2) also had a direct positive effect on the three facets of social skills (L1). Finally, classroom climate (L2) had an indirect positive effect on counterspeech (L1) via all three aspects of social skills (L1). CONCLUSION: The findings highlight that successful anti-hate speech programs may entail a combination of environmental and personal factors for increasing adolescents' active contribution to an inclusive and discrimination-free classroom environment where hate speech is not tolerated.

Roman Urdu Hate Speech Detection Using Transformer-Based Model for Cyber Security Applications.

Social media applications, such as Twitter and Facebook, allow users to communicate and share their thoughts, status updates, opinions, photographs, and videos around the globe. Unfortunately, some people utilize these platforms to disseminate hate speech and abusive language. The growth of hate speech may result in hate crimes, cyber violence, and substantial harm to cyberspace, physical security, and social safety. As a result, hate speech detection is a critical issue for both cyberspace and physical society, necessitating the development of a robust application capable of detecting and combating it in real-time. Hate speech detection is a context-dependent problem that requires context-aware mechanisms for resolution. In this study, we employed a transformer-based model for Roman Urdu hate speech classification due to its ability to capture the text context. In addition, we developed the first Roman Urdu pre-trained BERT model, which we named BERT-RU. For this purpose, we exploited the capabilities of BERT by training it from scratch on the largest Roman Urdu dataset consisting of 173,714 text messages. Traditional and deep learning models were used as baseline models, including LSTM, BiLSTM, BiLSTM + Attention Layer, and CNN. We also investigated the concept of transfer learning by using pre-trained BERT embeddings in conjunction with deep learning models. The performance of each model was evaluated in terms of accuracy, precision, recall, and F-measure. The generalization of each model was evaluated on a cross-domain dataset. The experimental results revealed that the transformer-based model, when directly applied to the classification task of the Roman Urdu hate speech, outperformed traditional machine learning, deep learning models, and pre-trained transformer-based models in terms of accuracy, precision, recall, and F-measure, with scores of 96.70%, 97.25%, 96.74%, and 97.89%, respectively. In addition, the transformer-based model exhibited superior generalization on a cross-domain dataset.

Effects of the Prevention Program "HateLess. Together against Hatred" on Adolescents' Empathy, Self-efficacy, and Countering Hate Speech.

Currently, there is a lack of empirically evaluated prevention programs targeting hate speech among adolescents. This is problematic because hate speech jeopardizes adolescents' well-being and social integration. To this end, this study aims to evaluate the short-term effects of the newly developed anti-hate speech prevention program, "HateLess. Together against Hatred", on adolescents' empathy, self-efficacy, and counter-speech. Eight hundred and twenty adolescents between 12 and 16 (M = 13.27, SD = 1.04) from 11 German schools participated in this study. More specifically, 567 adolescents participated in the one-week prevention program, and 253 participants were assigned to the control group. Repeated measures ANOVAs showed that HateLess was successful, as there was a significant increase in empathy, self-efficacy, and counter-speech in the intervention group from the pretest (T1) to the posttest (T2) one month after the intervention. In contrast, no changes were found among adolescents in the control group. A multilevel mediation model revealed that the effect of being a member of the intervention group on counter-speech was partially mediated via empathy and self-efficacy. The findings indicate that HateLess is an effective, cost-efficient approach to enhance adolescents' counter-speech directly and indirectly by altering the skills they need to become informed citizens in democratic societies.

Offensive, hateful comment: A networked discourse practice of blame and petition for justice during COVID-19 on Chinese Weibo.

Using data from user comments to the official social networking account of the Hubei Red Cross Foundation on a participatory web platform, this study attends to the offensive and hateful comments produced by ordinary Internet users to blame the elite authorities for their malfeasance in managing the donation during the COVID-19 in China. Drawing on Discursive Psychology, we focus on the rhetorical strategies that users employ to legitimise their actions as well-founded evidential blame against a norm-breaking act rather than radical extremist speech. The associated hatred among discussants are moral, social judgements. That said, hate speech also helps construct the moral standards of a normalised society.

"We're led by stupid people": Exploring Trump's use of denigrating and deprecating speech to promote hatred and violence.

In response to a call for criminologists to consider the impact of former President Donald Trump's presumed criminality, we analyze verbal-textual hostility (VTH) in Trump's campaign speeches. Politicians have particular power and reach with their speech and their use of VTH is an important part of the trifecta of violence. Using a framework informed by linguistic theory and previous analysis of hate speech in recorded hate crimes, we present the categories of deprecation and denigration, and discuss their relationship to domination. In context, these forms of VTH enhance and serve as precursors to more violent speech and acts.

Corporate Counterspeech.

Are corporations ever morally obligated to engage in counterspeech-that is, in speech that aims to counter hate speech and misinformation? While existing arguments in moral and political philosophy show that individuals and states have such obligations, it is an open question whether those arguments apply to corporations as well. In this essay, I show how two such arguments-one based on avoiding complicity, and one based on duties of rescue-can plausibly be extended to corporations. I also respond to several objections to corporate counterspeech.

Photovoice as Counterspeech.

A notorious hate group purchased anti-Muslim advertisements on buses operated by the San Francisco Municipal Transit Authority. The San Francisco Human Rights Commission engaged members of the Arab, Middle Eastern, Muslim, and South Asian communities in a photovoice project to explore the cultural identities, challenges, and resilience of community members coping with discrimination. The project provided a case example of photovoice as counterspeech and demonstrated the way in which counterspeech empowers affected communities to push back against harmful and threatening expression with resilience, cultural pride, and self-determination. Women and men in the photovoice participant group represented a wide range of backgrounds and ethnicities: Palestinian, Indian, Pakistani, and Lebanese. Religious affiliations included Muslim, Sikh, Christian, nondenominational, and agnostic. The exhibit was presented to the public in three major venues and was made available online.

Differences between antisemitic and non-antisemitic English language tweets.

Antisemitism is a global phenomenon on the rise that is negatively affecting Jews and communities more broadly. It has been argued that social media has opened up new opportunities for antisemites to disseminate material and organize. It is, therefore, necessary to get a picture of the scope and nature of antisemitism on social media. However, identifying antisemitic messages in large datasets is not trivial and more work is needed in this area. In this paper, we present and describe an annotated dataset that can be used to train tweet classifiers. We first explain how we created our dataset and approached identifying antisemitic content by experts. We then describe the annotated data, where 11% of conversations about Jews (January 2019-August 2020) and 13% of conversations about Israel (January-August 2020) were labeled antisemitic. Another important finding concerns lexical differences across queries and labels. We find that antisemitic content often relates to conspiracies of Jewish global dominance, the Middle East conflict, and the Holocaust.