People work to present themselves as moral, reasonable or justified even when making racist or hateful comments. In this project, we identify interactional practices that accomplish support for White nationalism as part of a reasonable (or even positive) identity held by reasonable or positive actors. Using membership categorisation analysis and conversation analysis, we analysed a corpus of 24 publicly-available video recordings for explicit mentions of (or challenges to) White nationalism/supremacy. Looking for how people explain, justify, and rationalise White nationalism and, especially, White nationalist violence, we identified 16 cases of what we call White nationalist remediation. Our findings demonstrate how not only self-avowed White nationalists but also those who do not publicly identify as such can work to protect and potentially normalise White nationalist views and actions, including violence, using the same practices. They thus (1) signal their followers, (2) present their beliefs as reasonable and defensible, and (3) ultimately normalise White nationalism.

Viral, victimized, and validated: A content analysis of supportive and hateful comments on LGBTQ+ TikTok.

Online violence and abuse pose significant challenges to public discourse, as it exacerbates existing power structures and marginalizes diverse epistemic perspectives. In this context, this study examines the epistemological consequences of hateful and toxic speech in online news comment sections, conceptualizing it as a form of epistemic violence—an effort to erase particular ways of thinking. Examining a dataset of toxic and hateful comments from The Conversation Canada, our findings emphasize four mechanisms of epistemic violence: insulting, labelling, ridiculing, and dehumanizing. These mechanisms function to delegitimize alternative epistemic positions and reinforce ideological conformity. Furthermore, these mechanisms disproportionately target those with marginalized identities along racial, gender, and political lines, further entrenching hegemonic power structures. Our research contributes to scholarship on digital epistemologies and platformized violence, highlighting the need for strategies that foster epistemic pluralism rather than simply suppressing toxic discourse.

Radical greens? how environmental influencers shape young social media users’ perceived environmental polarization, hopelessness, and collective action intentions

Background: In the 2024 election, Anies Baswedan used YouTube as a platform to live-stream his campaign titled ‘Desak Anies,’ with an open discussion. The campaign aimed to restore democracy and bring political discourse to the public. Purpose: This research mapped the interactions of netizens in the comment section of the live-streamed “Desak Anies” broadcast on his official YouTube channel. Methods: It employed thematic analysis. Data was collected using the ‘Google Apps Script’ tool, capturing YouTube comments from 13 episodes of ‘Desak Anies.’ The dataset included only direct comments, excluding replies, with a total of 6,996 comments. Result: The comments were mapped into three main themes: supportive comments, which were the most numerous; hateful comments, which were the fewest; and neutral. Conclusion: This study found a large number of supportive comments, with minimal opposition, indicating that “Desak Anies” strengthened Anies Baswedan’s existing support base but failed to attract new supporters. Thus, it creates the echo chamber effect and filters bubbles, preventing the campaign content from reaching broader viewers. Implications: Social media campaigns, such as “Desak Anies”, could potentially engage audiences. However, they often reinforce existing political beliefs, creating echo chambers that limit democratic discourse. Despite the campaign’s intention to foster political engagement, it strengthens existing support for Anies Baswedan and hinders the outreach to undecided or opposing voters. This suggests that digital platforms may inadvertently stifle true democratic deliberation by limiting exposure to diverse perspectives.

A multi-class cyberbullying classification on image and text in code-mixed Bangla-English social media content

This study delves into the linguistic and ideological dimensions of cyberbullying discourse directed at TikTok creator Putri Padang within the framework of Critical Discourse Analysis (CDA). In Indonesia’s digital landscape, TikTok has become one of the prominent platform for self-expression and cultural performance, yet it also serves as a site for public shaming and moral policing. Drawing on Fairclough’s (1995) three-dimensional model, this qualitative research analyzes fifty hate comments collected from several TikTok videos featuring Putri Padang to uncover how language reproduces power relations and cultural ideologies. The findings reveal that hate comments are not random acts of aggression but structured discursive practices characterized by repetition, labeling, and moral judgment. Linguistic strategies such as mockery, objectification, and intertextual humor—exemplified by terms like “muka kotak” and “Adudu”—function as mechanisms of symbolic domination, reinforcing gendered and regional hierarchies. Moreover, the comments often invoke patriarchal values and cultural authenticity to moral criticism, positioning the target as a violator of feminine and cultural norms. The research contributes to cyber-discourse studies by extending into multimodal contexts and emphasizing the need for culturally grounded approaches to online gender-based violence.

Content creators face heightened risks of online harm due to their public visibility and the increasing hostility in digital spaces. Traditional moderation tools, such as reporting and deletion, often fall short in addressing the emotional and reputational impact of online hate. In this study, we explore how community-driven approaches like counterspeech may support creators in navigating hateful comments, and how AI-powered tools could assist in this process. We conducted qualitative interviews and concept testing with 15 active content creators from Chinese social media platforms to understand their experiences and expectations. Our findings reveal that creators' approaches to counterspeech are shaped by their professional roles, the need to maintain authenticity, and the pressures of audience management. While creators saw value in AI-assisted counterspeech for improving efficiency and tone, they expressed concerns about misrepresentation, loss of control, and unintended harm. They preferred AI tools that augmented their voice rather than replaced it, and emphasized the importance of balancing emotional nuance with scalability. These insights inform the design of creator-centered AI moderation tools that integrate human agency and context sensitivity, contributing to healthier and more sustainable online environments.

This study investigates the dynamics of hate speech, looking at feedback on comments and subsequent commenting. We examine the relationship between feedback on comments, hate speech presence, and commenter types, with analysis of news comments during the 2022 South Korean presidential election campaigns. The data include 25 million comments, analyzed with a deep learning hate speech detection model. It was found that positive feedback encourages more commenting for non-hateful content, and negative feedback reduces subsequent non-hateful comments. However, surprisingly, negative feedback was found to rather increase the frequency of hateful comments, particularly among light commenters. Implications of the findings are discussed.

This study addresses three main questions: (1) What forms of illocutionary acts characterize hate speech in viral Instagram content involving religious figures? (2) How do these linguistic strategies polarize audience engagement in controversial religious discourse? (3) To what extent do platform mechanisms such as anonymity and algorithmic bias amplify toxicity? Through a qualitative netnographic analysis of five viral Instagram videos (late November–December 2023) featuring Gus Miftah and Ice Tea Seller three dominant hate speech themes emerge: religious-based attacks (52%), dehumanization (33%), and veiled threats (15%). Pragmatic analysis reveals that expressive illocutionary acts (60%—e.g., emotional outbursts) and directive acts (30%—e.g., demands for punishment) drive polarized engagement, with hate comments receiving three times as many likes and 35+ replies per thread compared to neutral comments. Platform dynamics exacerbate toxicity: 80% of hate comments come from anonymous accounts, while the algorithm promotes decontextualized clips, deepening ideological divisions. This study shows how linguistic aggression (micro-level) and platform architecture (macro-level) interact to normalize hate speech, offering actionable strategies for creators to counter hostility (e.g., context restoration) and platforms to prioritize ethical algorithms. By integrating linguistic theory with digital ethics, this study advances a framework for mitigating harm in Indonesia’s polarized social media landscape.

Despite ongoing efforts, online hate speech remains a pervasive issue on social media, particularly affecting vulnerable groups such as LGBTQ communities. While there is extensive debate around how best to address this problem, counter speech is emerging as a promising solution. However, existing research has primarily focused on detecting hateful content, often overlooking broader aspects such as the specific topics of discrimination and the spread of countermeasures online. This study examines the prevalence of hate speech and counter speech in LGBTQ online spaces on TikTok, analysing day-to-day interactions to identify recurring themes and targets. Results reveal that hate speech is widespread: at least 3.5% of messages contain hateful content, spread by approximately 4% of users, and one in three videos attracts hate comments or replies, primarily targeting LGBTQ topics explicitly. Gender identity emerges as a major focus, with transgender and non-binary individuals being frequent targets. Although much hate engagement goes unanswered, when responses occur, they are often in the form of counter speech, especially when LGBTQ-related topics are targeted. These findings improve our understanding of the nature and extent of online hate speech against LGBTQ communities, confirm counter speech as an employed response, and provide a foundation for further research aimed at developing strategies to promote safer, more inclusive social media environments.

ABSTRACT In this study, we investigated the prevalence and nature of hate speech in comments on the posts of Pakistani political journalists on X (formerly Twitter). We applied the mob censorship theory to analyze the hatred spread by commenters aiming to intimidate critical journalists. Our findings indicate that both male and female journalists received an unprecedented number of hateful comments, including life-threatening messages and highly dehumanizing and profane expressions. As anticipated, female journalists received a larger volume of hateful comments, primarily targeting their gender rather than their profession. Similarly, political posts attracted a higher number of negative comments compared to nonpolitical ones. We argue that the significant volume of online hate, mainly originating from the political sphere, was further fueled by existing socio-cultural practices. This escalation can render individuals more vulnerable to mob censorship and deter them from expressing their opinions independently on social media platforms.

Everyone has the right to freedom of expression. However, under the guise of free speech, this privilege is being abused to discriminate against and harm others, either physically or verbally. Hate speech is the term for this type of bigotry. Hate speech is described as language used to show hatred toward an individual or a group of individuals based on traits such as race, religion, ethnicity, gender, nationality, handicap, and sexual orientation. It can take the form of speech, writing, gestures, or displays that target someone due to their affiliation with a particular group. Hate speech has been more prevalent in recent years, both in person and online. Hateful content is bred and shared on social media and other internet platforms, which finally leads to hate crimes. The growing use of social media platforms and information exchange has resulted in significant benefits for humanity. However, this has resulted in several issues, including the spread and dissemination of hate speech messages. Recent studies used a range of machine learning and deep learning techniques with text mining methods to automatically detect hate speech messages on real-time datasets to handle this developing issue on social media platforms. Hence, this paper aims to survey the various algorithms to detect hateful comments and predict the best algorithms in social media datasets. And also implemented in real-time social environments to detect hate speech with mobile intimation.

This study aims to analyze the impact of hate comments on Instagram on the self-confidence of students from the Faculty of Social Sciences (FIS) at Universitas Islam Negeri Sumatera Utara (UINSU), class of 2021. Using a descriptive qualitative method, the research identifies psychological impacts experienced by students, such as anxiety, stress, and low self-esteem. The study also highlights strategies that students can employ to mitigate these impacts, including digital ethics education, digital literacy, and social support. The findings reveal that hate comments affect not only individual psychological conditions but also their social relationships. To address this issue, a holistic approach is required, including enhancing digital literacy and implementing protective policies on social media platforms.

온라인 뉴스 댓글 공간에서의 혐오 표현의 만연은 심각한 사회적 문제로 대두되고 있다. 본 연구는 국내 주요 포털인 네이버의 뉴스 댓글 109,126개를 수집하여 혐오 분류 모델 '언스마일(Unsmile)'과 토픽 모델링 기법 LDA를 활용한 분석을 수행하였다. 분석 결과, 정치, 사회, 세계 분야에서 혐오 표현이 전체 댓글 비중에 40% 이상으로 나타났으며, 이는 현행 댓글 인터페이스의 한계점을 드러내는 것으로 확인되었다. 위 분석 결과를 토대로, 본 연구는 댓글의 혐오 수준을 판별하고 시각화하는 인터페이스를 설계하였다. 제안된 인터페이스의 효과성을 검증하기 위해 280명의 사용자를 대상으로 통계 분석을 실시하였으며, 그 결과 해당 인터페이스가 혐오 표현 감소와 댓글 기피 성향 완화에 유의미한 효과가 있음을 확인하였다. 본 연구는 국내 포털의 혐오 표현 실태를 실증적으로 분석하고, 인터페이스 개선안의 효과와 한계점을 제시했다는 점에서 학술적 의의를 지닌다.

This research aims to (1) identify the impoliteness strategies in hate speech comments on Noah Schnapp’s Instagram posts and (2) explain the function of impoliteness strategies manifested by netizens in Noah Schnapp’s Instagram posts. The controversy that arose after Schnapp openly supported Zionism triggered a flood of hate comments from netizens. The research used Jonathan Culpeper’s theory of impoliteness to identify five types of impoliteness strategies: negative impoliteness, bald on record impoliteness, positive impoliteness, sarcasm or mock politeness, and withhold politeness. Data were collected qualitatively and analyzed descriptively. As a result, out of 30 data analyzed, the negative impoliteness strategy was the most dominant (15 data), followed by bald on record (7 data), sarcasm or mock politeness (4 data), and positive impoliteness (4 data), while withhold politeness was not found. In terms of function, most of the comments reflect affective functions (16 data), which show emotional outbursts such as anger or disappointment, followed by coercive functions (9 data), and entertainment functions (5 data). This research provides insight into how netizens use impoliteness strategies to attack celebrities' public images on social media.

In the digital era, social media platforms have seen a substantial increase in the volume of online comments. While these platforms provide users with a space to express their opinions, they also serve as fertile ground for the proliferation of hate speech. Hate comments can be categorized into various types, including discrimination, violence, racism, and sexism, all of which can negatively impact mental health. Among these, sexism poses a significant challenge due to its various forms and the difficulty in defining it, making detection complex. Nevertheless, detecting and preventing sexism on social networks remains a critical issue. Recent studies have leveraged language models such as transformers, known for their ability to capture the semantic nuances of textual data. In this study, we explore different transformer models, including multiple versions of RoBERTa (A Robustly Optimized BERT Pretraining Approach), to detect sexism. We hypothesize that combining a sentiment-focused language model with models specialized in sexism detection can improve overall performance. To test this hypothesis, we developed two approaches. The first involved using classical transformers trained on our dataset, while the second combined embeddings generated by transformers with a Long Short-Term Memory (LSTM) model for classification. The probabilistic outputs of each approach were aggregated through various voting strategies to enhance detection accuracy. The LSTM with embeddings approach improved the F1-score by 0.2% compared to the classical transformer approach. Furthermore, the combination of both approaches confirms our hypothesis, achieving a 1.6% improvement in the F1-score in each case. We determined that an F1 score of over 0.84 effectively measures sexism. Additionally, we constructed our own dataset to train and evaluate the models.

Malicious comments from netizens circulating on social media in Kupang City are increasingly frequent. Over the past five years, there have been 225 police reports regarding malicious comments from netizens whose content leads to criminal acts of insult and/or defamation and hate speech, but the handling by the police in the Kupang City Police Resort area has not been optimal. Therefore, the main problem of this study is what are the inhibiting factors in handling malicious comments from netizens and the efforts made by the police in the Kupang City Police Resort area to overcome these inhibiting factors. The purpose of this study is to determine and describe the inhibiting factors and efforts made by the police in the Kupang City Police Resort area in handling malicious comments from netizens in Kupang City. The methods used are document/literature studies and field research by interviewing a number of key informants. The findings of this study show that the inhibiting factors in handling malicious comments from netizens by the police in the Kupang City Police Resort are (1) limited number, quality and capability of police personnel who are authorized and tasked with handling cybercrime, including malicious comments from netizens; (2) limited supporting facilities and infrastructure, both quantitatively and qualitatively, such as equipment and finances; (3) lack of knowledge and understanding of laws and regulations regarding netizen hate comments.

The rise in abusive language on social media is a significant threat to mental health and social cohesion. For Bengali speakers, the need for effective detection is critical. However, current methods fall short in addressing the massive volume of content. Improved techniques are urgently needed to combat online hate speech in Bengali. Traditional machine learning techniques, while useful, often require large, linguistically diverse datasets to train models effectively. This paper addresses the urgent need for improved hate speech detection methods in Bengali, aiming to fill the existing research gap. Contextual understanding is crucial in differentiating between harmful speech and benign expressions. Large language models (LLMs) have shown state-of-the-art performance in various natural language tasks due to their extensive training on vast amounts of data. We explore the application of LLMs, specifically GPT-3.5 Turbo and Gemini 1.5 Pro, for Bengali hate speech detection using Zero-Shot and Few-Shot Learning approaches. Unlike conventional methods, Zero-Shot Learning identifies hate speech without task-specific training data, making it highly adaptable to new datasets and languages. Few-Shot Learning, on the other hand, requires minimal labeled examples, allowing for efficient model training with limited resources. Our experimental results show that LLMs outperform traditional approaches. In this study, we evaluate GPT-3.5 Turbo and Gemini 1.5 Pro on multiple datasets. To further enhance our study, we consider the distribution of comments in different datasets and the challenge of class imbalance, which can affect model performance. The BD-SHS dataset consists of 35,197 comments in the training set, 7542 in the validation set, and 7542 in the test set. The Bengali Hate Speech Dataset v1.0 and v2.0 include comments distributed across various hate categories: personal hate (629), political hate (1771), religious hate (502), geopolitical hate (1179), and gender abusive hate (316). The Bengali Hate Dataset comprises 7500 non-hate and 7500 hate comments. GPT-3.5 Turbo achieved impressive results with 97.33%, 98.42%, and 98.53% accuracy. In contrast, Gemini 1.5 Pro showed lower performance across all datasets. Specifically, GPT-3.5 Turbo excelled with significantly higher accuracy compared to Gemini 1.5 Pro. These outcomes highlight a 6.28% increase in accuracy compared to traditional methods, which achieved 92.25%. Our research contributes to the growing body of literature on LLM applications in natural language processing, particularly in the context of low-resource languages.

In the last decade, the world has witnessed remarkable technological development, especially in artificial intelligence, which helps researchers find solutions to problems of concern to the individual and society, mainly, the huge propagation of hate speech with the increased use of social media platforms. In this study, we aim to enhance the detection of Arabic hate speech on social media by addressing challenges related to imbalanced datasets through data augmentation techniques. Several machine learning algorithms and the DziriBert, a pre-trained transformer model, are implemented on the Tunisian Hate Speech and Abusive Dataset (T-HSAB). The proposed approach achieves good results, improving the detection of hateful comments on Arabic social media using the Synthetic Minority Over-sampling Technique (SMOTE). Notably, the DziriBert model exhibits remarkable proficiency in detecting hate speech, achieving an accuracy of 82%. Random Forest (RF) and Linear SVC outperform the state of the art approaches, achieving the best result.