Offensive Comments Research Articles

University Students’ Perceptions and Reactions to Cyberbullying on Social Media: The Case of Türkiye

ABSTRACT With cyberbullying emerging as a significant concern in contemporary society, understanding its prevalence, forms, and impacts is crucial for developing effective intervention strategies. This study aims to delve into the underlying factors contributing to cyberbullying incidents among university students. Employing quantitative methodologies, this research utilized a questionnaire comprising multiple-choice, scaled responses, and open-ended questions to explore students’ encounters, perceptions, and responses to cyberbullying. The study’s development was informed by theoretical frameworks and existing literature, with data analysis conducted using the Statistical Package for the Social Sciences (SPSS). The sample comprised 392 participants, predominantly female (79.1 percent) and Turkish (98.7 percent). Most participants were 18–21 (46.9 percent) and 21–25 (41.3 percent). The majority (84.9 percent) were studying at a state university. The study reveals a widespread acknowledgment of cyberbullying as a severe problem among social media users, particularly prevalent on platforms like Instagram and TikTok. Offensive comments and hate speech were identified as dominant forms of cyberbullying, alongside newer tactics like emoji-based ridicule. Participant responses ranged from passive bystander behavior to active intervention, reflecting diverse coping mechanisms and strategies. The findings underscore the need for comprehensive approaches to combat cyberbullying, including educational programs, institutional support mechanisms, and interventions promoting digital citizenship. Furthermore, fostering peer support networks and enhancing awareness in university settings is crucial for effectively addressing cyberbullying. The study highlights various intervention areas and approaches to mitigate the impact of cyberbullying and promote a healthier digital environment. Future research should continue exploring the underlying motivations and barriers to seeking help, aiming to develop targeted strategies for prevention and intervention.

Hate speech detection in low-resourced Indian languages: An analysis of transformer-based monolingual and multilingual models with cross-lingual experiments

Abstract Warning: This paper is based on hate speech detection and may contain examples of abusive/ offensive phrases. Cyberbullying, online harassment, etc., via offensive comments are pervasive across different social media platforms like ™Twitter, ™Facebook, ™YouTube, etc. Hateful comments must be detected and eradicated to prevent harassment and violence on social media. In the Natural Language Processing (NLP) domain, the most prevalent task is comment classification, which is challenging, and language models based on transformers are at the forefront of this advancement. This paper intends to analyze the performance of language models based on transformers like BERT, ALBERT, RoBERTa, and DistilBERT on the Indian hate speech datasets over binary classification. Here, we utilize the existing datasets, i.e., HASOC (Hindi and Marathi) and HS-Bangla. So, we evaluate several multilingual language models like MuRIL-BERT, XLM-RoBERTa, etc., few monolingual language models like RoBERTa-Hindi, Maha-BERT (Marathi), Bangla-BERT (Bangla), Assamese-BERT (Assamese), etc., and perform cross-lingual experiment also. For further analyses, we perform multilingual, monolingual, and cross-lingual experiments on our Hate Speech Assamese (HS-Assamese) (Indo-Aryan language family) and Hate Speech Bodo (HS-Bodo) (Sino-Tibetan language family) dataset (HS dataset version 2) also and achieved a promising result. The motivation of the cross-lingual experiment is to encourage researchers to learn about the power of the transformer. Note that no pre-trained language models are currently available for Bodo or any other Sino-Tibetan languages.

DÉDUCTION EFFICACE DE LANGUE OFFENSIVE UTILISANT L'APPRENTISSAGE PROFONDE DANS LES MÉDIAS SOCIAUX

Offensive language detection is the technique of identifying and detecting user-generated offensive comments such as insults, pain, profanity, and racism that are targeted at a specific individual or group on social media. As social media platforms become more prominent, offensive language is used more frequently, becoming a major challenge in modern society. A novel effective offensive language classification (EOLC) technique has been proposed to overcome these challenges. English language tweets from YouTube and X (Twitter) with offensive, mild, swear, and non-offensive tweets are used in this paper. Initially, the tweets and comments are pre-processed, and the features are extracted using different techniques, namely term frequency-inverse document frequency (TF-IDF), WordVec, and lexicon-based features. The extracted features are classified using the graph-based deep learning (GDL) method for numerical representation and decision-making. GDL network is optimized with red fox optimization (RFO) to normalize the weight and biases of the network and achieve better accuracy. The proposed GDL model achieves the highest levels of classification accuracy on the X (Twitter) and YouTube datasets, with 95.5 % and 96.8 %, respectively. The results obtained from GDL are more accurate and of higher quality than those obtained from traditional classifiers. The proposed EOLC method improves the overall accuracy by 5.56 %, 7.4 %, 7.7 %, and 10.2 % better than Text CNN, CNN-LSTM, DRNN, and LogitBoost, respectively.

Detecting Offensive Language on Malay Social Media: A Zero-Shot, Cross-Language Transfer Approach Using Dual-Branch mBERT

Social media serves as a platform for netizens to stay informed and express their opinions through the Internet. Currently, the social media discourse environment faces a significant security threat—offensive comments. A group of users posts comments that are provocative, discriminatory, and objectionable, intending to disrupt online discussions, provoke others, and incite intergroup conflict. These comments undermine citizens’ legitimate rights, disrupt social order, and may even lead to real-world violent incidents. However, current automatic detection of offensive language primarily focuses on a few high-resource languages, leaving low-resource languages, such as Malay, with insufficient annotated corpora for effective detection. To address this, we propose a zero-shot, cross-language unsupervised offensive language detection (OLD) method using a dual-branch mBERT transfer approach. Firstly, using the multi-language BERT (mBERT) model as the foundational language model, the first network branch automatically extracts features from both source and target domain data. Subsequently, Sinkhorn distance is employed to measure the discrepancy between the source and target language feature representations. By estimating the Sinkhorn distance between the labeled source language (e.g., English) and the unlabeled target language (e.g., Malay) feature representations, the method minimizes the Sinkhorn distance adversarially to provide more stable gradients, thereby extracting effective domain-shared features. Finally, offensive pivot words from the source and target language training sets are identified. These pivot words are then removed from the training data in a second network branch, which employs the same architecture. This process constructs an auxiliary OLD task. By concealing offensive pivot words in the training data, the model reduces overfitting and enhances robustness to the target language. In the end-to-end framework training, the combination of cross-lingual shared features and independent features culminates in unsupervised detection of offensive speech in the target language. The experimental results demonstrate that employing cross-language model transfer learning can achieve unsupervised detection of offensive content in low-resource languages. The number of labeled samples in the source language is positively correlated with transfer performance, and a greater similarity between the source and target languages leads to better transfer effects. The proposed method achieves the best performance in OLD on the Malay dataset, achieving an F1 score of 80.7%. It accurately identifies features of offensive speech, such as sarcasm, mockery, and implicit expressions, and showcases strong generalization and excellent stability across different target languages.

Profile update: the effects of identity disclosure on network connections and language

Our social identities determine how we interact and engage with the world surrounding us. In online settings, individuals can make these identities explicit by including them in their public biography, possibly signaling a change in what is important to them and how they should be viewed. While there is evidence suggesting the impact of intentional identity disclosure in online social platforms, its actual effect on engagement activities at the user level has yet to be explored. Here, we perform the first large-scale study on Twitter that examines behavioral changes following identity disclosure on Twitter profiles. Combining social networks with methods from natural language processing and quasi-experimental analyses, we discover that after disclosing an identity on their profiles, users (1) tweet and retweet more in a way that aligns with their respective identities, and (2) connect more with users that disclose similar identities. We also examine whether disclosing the identity increases the chance of being targeted for offensive comments and find that in fact (3) the combined effect of disclosing identity via both tweets and profiles is associated with a reduced number of offensive replies from others. Our findings highlight that the decision to disclose one’s identity in online spaces can lead to substantial changes in how they express themselves or forge connections, with a lesser degree of negative consequences than anticipated.

Correction: HATDO: hybrid Archimedes Tasmanian devil optimization CNN for classifying offensive comments and non-offensive comments

Correction: HATDO: hybrid Archimedes Tasmanian devil optimization CNN for classifying offensive comments and non-offensive comments

“HOT” ChatGPT: The Promise of ChatGPT in Detecting and Discriminating Hateful, Offensive, and Toxic Comments on Social Media

Harmful textual content is pervasive on social media, poisoning online communities and negatively impacting participation. A common approach to this issue is developing detection models that rely on human annotations. However, the tasks required to build such models expose annotators to harmful and offensive content and may require significant time and cost to complete. Generative AI models have the potential to understand and detect harmful textual content. We used ChatGPT to investigate this potential and compared its performance with MTurker annotations for three frequently discussed concepts related to harmful textual content on social media: Hateful, Offensive, and Toxic (HOT). We designed five prompts to interact with ChatGPT and conducted four experiments eliciting HOT classifications. Our results show that ChatGPT can achieve an accuracy of approximately 80% when compared to MTurker annotations. Specifically, the model displays a more consistent classification for non-HOT comments than HOT comments compared to human annotations. Our findings also suggest that ChatGPT classifications align with the provided HOT definitions. However, ChatGPT classifies “hateful” and “offensive” as subsets of “toxic.” Moreover, the choice of prompts used to interact with ChatGPT impacts its performance. Based on these insights, our study provides several meaningful implications for employing ChatGPT to detect HOT content, particularly regarding the reliability and consistency of its performance, its understanding and reasoning of the HOT concept, and the impact of prompts on its performance. Overall, our study provides guidance on the potential of using generative AI models for moderating large volumes of user-generated textual content on social media.

Recognizing and Mitigating Racial and Ethnic Interview-Associated Microaggressions Across the Academy

The impact of microaggressions in pharmacy education interviews reinforces a message of assimilation for acceptance rather than inviting a potential underrepresented minority (URM) colleague’s authentic identity they could bring to diversify and strengthen the institution. Interviewers may ask inappropriate questions or make offensive comments without intending to cause harm; however, the impact of the words on the interviewee are valid and cannot be ignored. This text focuses on the challenges faced by URM faculty in the context of academic pharmacy interviews which potentially contribute to their underrepresentation in academic pharmacy. The authors provide examples of perspectives and experiences from URM pharmacy faculty when interviewing for academic pharmacy positions. This text proposes several recommendations such as microaggression prevention strategies, bias training, and institutional culture improvement. These solutions can help institutions achieve a sustainable means of recruiting and retaining URM faculty.

Exploring the occurrence of microaggressions in the genetic counseling student-supervisor relationship: A mixed-methods study.

While research has shown that genetic counseling students with minoritized racial or ethnic identities face microaggressions throughout graduate training, quantitative data regarding the frequency of these experiences have not been reported. The purpose of this mixed-methods study was to investigate the frequency and types of microaggressions experienced by graduates of accredited genetic counseling programs in the United States during fieldwork rotations. A quantitative survey was administered to assess how frequently 14 different types of microaggressions occurred in interactions with supervisors. Survey responses were analyzed using situation-based coding (the number of different types of microaggressions experienced) and frequency-based coding (the sum of participants' weighted Likert answers). Select survey respondents with minoritized identities were interviewed to better contextualize and categorize microaggression experiences. Analysis of 87 survey responses revealed that participants with minoritized racial and ethnic identities experience significantly more types of microaggressions (t(61) = 2.77; p = 0.007) at a significantly higher frequency (t(55) = 2.67; p = 0.010) than their white counterparts. Participants who identified as part of the disability community were also found to experience significantly more types of microaggressions (t(10) = 3.25; p = 0.009) at a significantly higher frequency (t(9) = 2.32; p = 0.045) than those who did not. Qualitative analysis of 11 interviews revealed that microaggressions from supervisors included offensive and inappropriate comments, unequal treatment, cultural intolerance, and disparaging feedback. Overall, our data present evidence that students with minoritized racial and ethnic identities and students with disabilities are subjected to a variety of inequitable, exclusionary, and harmful interactions. As a result, we recommend that all supervisors receive training about recognizing and preventing microaggressions to ensure that students are provided with an equitable and inclusive training experience, regardless of identity.

A Study on Chinese Social Media Xiaohongshu Users' Attitudes Toward Offensive Comments

With the rapid development of emerging digital technologies, social media platforms have proliferated like mushrooms after rain. Xiaohongshu is among the highly trusted "word-of-mouth libraries" for Chinese social media users, offering numerous conveniences while also subjecting users to disruptive offensive language. This study analyzes the various emotions and behavioral mechanisms exhibited by Xiaohongshu users when confronted with offensive language, as well as the underlying reasons for these reactions. It underscores the positive impact on society of enhancing online supervision, safeguarding user privacy and security, optimizing service functionalities on social media platforms, increasing user engagement, and fostering media literacy among users to promote moderate use of social media. Through questionnaire surveys, the study reveals that offensive language on Xiaohongshu induces negative emotions such as anger, sadness, confusion, and anxiety among users. Consequently, users employ strategies such as rebuttal, reporting, and even reducing their frequency of using Xiaohongshu to counteract the harm caused by offensive language.

Users' Experiences With Online Access to Electronic Health Records in Mental and Somatic Health Care: Cross-Sectional Study.

Patient-accessible electronic health records (PAEHRs) hold promise for empowering patients, but their impact may vary between mental and somatic health care. Medical professionals and ethicists have expressed concerns about the potential challenges of PAEHRs for patients, especially those receiving mental health care. This study aims to investigate variations in the experiences of online access to electronic health records (EHRs) among persons receiving mental and somatic health care, as well as to understand how these experiences and perceptions vary among those receiving mental health care at different levels of point of care. Using Norwegian data from the NORDeHEALTH 2022 Patient Survey, we conducted a cross-sectional descriptive analysis of service use and perceptions of perceived mistakes, omissions, and offensive comments by mental and somatic health care respondents. Content analysis was used to analyze free-text responses to understand how respondents experienced the most serious errors in their EHR. Among 9505 survey participants, we identified 2008 mental health care respondents and 7086 somatic health care respondents. A higher percentage of mental health care respondents (1385/2008, 68.97%) reported that using PAEHR increased their trust in health care professionals compared with somatic health care respondents (4251/7086, 59.99%). However, a significantly larger proportion (P<.001) of mental health care respondents (976/2008, 48.61%) reported perceiving errors in their EHR compared with somatic health care respondents (1893/7086, 26.71%). Mental health care respondents also reported significantly higher odds (P<.001) of identifying omissions (758/2008, 37.75%) and offensive comments (729/2008, 36.3%) in their EHR compared with the somatic health care group (1867/7086, 26.35% and 826/7086, 11.66%, respectively). Mental health care respondents in hospital inpatient settings were more likely to identify errors (398/588, 67.7%; P<.001) and omissions (251/588, 42.7%; P<.001) than those in outpatient care (errors: 422/837, 50.4% and omissions: 336/837, 40.1%; P<.001) and primary care (errors: 32/100, 32% and omissions: 29/100, 29%; P<.001). Hospital inpatients also reported feeling more offended (344/588, 58.5%; P<.001) by certain content in their EHR compared with respondents in primary (21/100, 21%) and outpatient care (287/837, 34.3%) settings. Our qualitative findings showed that both mental and somatic health care respondents identified the most serious errors in their EHR in terms of medical history, communication, diagnosis, and medication. Most mental and somatic health care respondents showed a positive attitude toward PAEHRs. However, mental health care respondents, especially those with severe and chronic concerns, expressed a more critical attitude toward certain content in their EHR compared with somatic health care respondents. A PAEHR can provide valuable information and foster trust, but it requires careful attention to the use of clinical terminology to ensure accurate, nonjudgmental documentation, especially for persons belonging to health care groups with unique sensitivities.

Transformer based model for offensive content recognition in dravidian languages

This paper describes a model for spotting offensive data from the comments being collected from social media. The comments posted will include expressions, emoticons and will mostly be in code mixed language and classifying these code-mixed language comments is tricky. The proposed system uses a multi-head attention model to extract features from the code-mixed Tamil input data. Various classification algorithms are applied to these extracted features to categorize offensive comments. The generated labels are optimized by performing majority voting on labels generated by different algorithms. This system is validated on the validation set and is evaluated by applying the Tamil CodeMix test data from the dataset published by the HASOC task (Task2-subtask1) at FIRE 2021. The evaluation yields an average weighted F1 score of 0.83 and is ranked 3rd position in the official ranking.

XBert - A Model for Hate Speech Detection in Vietnamese Text

— In the digital age, social media's pervasive influence has inadvertently escalated the prevalence of hate speech and offensive comments, with alarming implications for mental health. There is increasing evidence indicating a clear correlation between two factors. exposure to such toxic online content and the onset of depression among users, particularly affecting vulnerable groups like content creators and channel owners. Addressing this critical issue, our research introduces XBert, a model for detecting hostile and provocative language in Vietnamese. We propose an approach related to data preprocessing, improved tokenization, and model fine-tuning. We have modified the architecture of the Roberta model, used the EDA technique, and added a dropout parameter to the tokenizer. Our model achieved an accuracy of 99.75% and an F1-Macro score of 98.05%. This is a promising result for a model detecting provocative and hostile language in Vietnamese.

Online Harassment and Trolling of Political Journalists in Pakistan

ABSTRACT This study examines the online harassment faced by political journalists in Pakistan on Twitter, specifically focusing on the actions of users affiliated with various political parties. By selecting the 12 most active journalists on Twitter, we combined both content and textual analyses to examine comments posted on their tweets. Drawing upon the theoretical framework of the practice-based theoretical approach, the findings reveal that these journalists predominantly encountered negative comments encompassing personal abuses, culturally sensitive expressions, and attacks on their professional lives. Commenters associated with the populist political party, Pakistan Tehreek Insaaf, were more likely to engage in uncivil comments compared to those affiliated with traditional political parties. Interestingly, both male and female journalists received a comparable number of comments on their tweets; however, the nature of the comments differed significantly. Male journalists were primarily criticized for their perceived lack of professionalism, whereas female journalists were subjected to gender-targeted offensive comments. The study highlighted how socio-cultural conditions were exploited by political parties and netizens to suppress critical journalists. Overall, this research sheds light on the multifaceted dynamics of harassment faced by journalists in the digital age, emphasizing the need for effective measures to safeguard freedom of expression and counteract online abuse within the Pakistani media landscape.

Toxic Comment Classification using Deep Learning

Online Conversation media serves as a means for individuals to engage, cooperate, and exchange ideas; however, it is also considered a platform that facilitates the spread of hateful and offensive comments, which could significantly impact one's emotional and mental health. The rapid growth of online communication makes it impractical to manually identify and filter out hateful tweets. Consequently, there is a pressing need for a method or strategy to eliminate toxic and abusive comments and ensure the safety and cleanliness of social media platforms. Utilizing LSTM, Character-level CNN, Word-level CNN, and Hybrid model (LSTM + CNN) in this toxicity analysis is to classify comments and identify the different types of toxic classes by means of a comparative analysis of various models. The neural network models utilized for this analysis take in comments extracted from online platforms, including both toxic and non-toxic comments. The results of this study can contribute towards the development of a web interface that enables the identification of toxic and hateful comments within a given sentence or phrase, and categorizes them into their respective toxicity classes.

XRBi-GAC: A hybrid deep learning framework for multilingual toxicity detection

Social media platforms allow people across the globe to share their thoughts and opinions and conveniently communicate with each other. Apart from various advantages of social media, it is also misused by a set of users for hate-mongering with toxic and offensive comments. The majority of the earlier proposed toxicity detection methods are primarily focused on the English language, but there is a lack of research on low-resource languages and multilingual text data. We propose an XRBi-GAC framework comprising XLM-RoBERTa, Bi-GRU with self-attention and capsule networks for multilingual toxic text detection. A loss function is also presented, which fuses the binary cross-entropy loss and focal loss to address the class imbalance problem. We evaluated the proposed framework on two datasets, namely, the Jigsaw Multilingual Toxic Comment dataset and HASOC 2019 dataset and achieved F1-score of 0.865 and 0.829, respectively. The results of the experiments show that the proposed framework has outperformed the state-of-the-art multilingual models XLM-RoBERTa and mBERT on both datasets, which shows the versatility and robustness of the proposed XRBi-GAC framework.

Inequitable treatment as perceived by international medical graduates (IMGs): a scoping review

ObjectivesThis scoping review seeks to detail experiences of inequitable treatment, as self-reported by international medical graduates (IMGs), across time and location.DesignScoping review.Search strategyThree academic medical databases (MEDLINE, SCOPUS and PSYCINFO)...

HATDO: hybrid Archimedes Tasmanian devil optimization CNN for classifying offensive comments and non-offensive comments

HATDO: hybrid Archimedes Tasmanian devil optimization CNN for classifying offensive comments and non-offensive comments

OLID-BR: offensive language identification dataset for Brazilian Portuguese

Social media has revolutionized the manner in which our society is interconnected. While this extensive connectivity offers numerous benefits, it is also accompanied by significant drawbacks, particularly in terms of the proliferation of fake news and the vast dissemination of hate speech. Identifying offensive comments is a critical task for ensuring the safety of users, which is why industry and academia have been working on developing solutions to this problem. Prior research on hate speech detection has predominantly focused on the English language, with few studies devoted to other languages such as Portuguese. This paper introduces the Offensive Language Identification Dataset for Brazilian Portuguese (OLID-BR), a high-quality NLP dataset for offensive language detection, which we make publicly available. The dataset contains 6,354 (extendable to 13,538) comments labeled using a fine-grained three-layer annotation schema compatible with datasets in other languages, which allows the training of multilingual/cross-lingual models. The five NLP tasks available in OLID-BR allow the detection of offensive comments, the classification of the types of offenses such as racism, LGBTQphobia, sexism, xenophobia, and so on, the identification of the type and the target of offensive comments, and the extraction of toxic spans of offensive comments. All those tasks can enhance the capabilities of content moderation systems by providing deep contextual analysis or highlighting the spans that make a text toxic. We further experiment with and evaluate the dataset using state-of-the-art BERT-based and NER models, which demonstrates the usefulness of OLID-BR for the development of toxicity detection systems for Portuguese texts.

Classification of Virtual Harassment on Social Networks Using Ensemble Learning Techniques

Background: Internet social media platforms have become quite popular, enabling a wide range of online users to stay in touch with their friends and relatives wherever they are at any time. This has led to a significant increase in virtual crime from the inception of these platforms to the present day. Users are harassed online when confidential information about them is stolen, or when another user posts insulting or offensive comments about them. This has posed a significant threat to online social media users, both mentally and psychologically. Methods: This research compares traditional classifiers and ensemble learning in classifying virtual harassment in online social media networks by using both models with four different datasets: seven machine learning algorithms (Nave Bayes NB, Decision Tree DT, K Nearest Neighbor KNN, Logistics Regression LR, Neural Network NN, Quadratic Discriminant Analysis QDA, and Support Vector Machine SVM) and four ensemble learning models (Ada Boosting, Gradient Boosting, Random Forest, and Max Voting). Finally, we compared our results using twelve evaluation metrics, namely: Accuracy, Precision, Recall, F1-measure, Specificity, Matthew’s Correlation Coefficient (MCC), Cohen’s Kappa Coefficient KAPPA, Area Under Curve (AUC), False Discovery Rate (FDR), False Negative Rate (FNR), False Positive Rate (FPR), and Negative Predictive Value (NPV) were used to show the validity of our algorithms. Results: At the end of the experiments, For Dataset 1, Logistics Regression had the highest accuracy of 0.6923 for machine learning algorithms, while Max Voting Ensemble had the highest accuracy of 0.7047. For dataset 2, K-Nearest Neighbor, Support Vector Machine, and Logistics Regression all had the same highest accuracy of 0.8769 in the machine learning algorithm, while Random Forest and Gradient Boosting Ensemble both had the highest accuracy of 0.8779. For dataset 3, the Support Vector Machine had the highest accuracy of 0.9243 for the machine learning algorithms, while the Random Forest ensemble had the highest accuracy of 0.9258. For dataset 4, the Support Vector Machine and Logistics Regression both had 0.8383, while the Max voting ensemble obtained an accuracy of 0.8280. A bar chart was used to represent our results, showing the minimum, maximum, and quartile ranges. Conclusions: Undoubtedly, this technique has assisted in no small measure in comparing the selected machine learning algorithms as well as the ensemble for detecting and exposing various forms of cyber harassment in cyberspace. Finally, the best and weakest algorithms were revealed.