Hate Tweets Research Articles

Use of Natural Language Processing for the Detection of Hate Speech on Social Media

Our society’s communication patterns have fundamentally changed as a consequence of the emergence of social media platforms. One effect of these changes is a rise in unpleasant behaviours like making rude and derogatory comments online. Speaking harshly or disrespectfully to someone in person may be difficult. However, online abuse and posting of improper material are considered to be acceptable. Hate speech has the potential to hurt a person or a group of people. Inappropriate material must be identified, in order to be filtered or banned from the web. CNN is a type of deep machine-learning model that has been suggested for such identification, because it performs better than conventional techniques in resolving text categorization problems. Our goal investigates how hate speech may be detected using NLP. In addition, a recent technique has been used in this field to a dataset. This classifier is assigned in each tweet to one of the three Twitter dataset categories of hatred, foul language, or neither. This model’s performance has been assessed with accuracy. The Naïve Bayes, the Decision Tree, KNN, Linear Regression, and the Random Forest are five algorithms that have been used. Of these, Linear Regression provided the greatest accuracy of 94%. It should be noted that when looking at each class separately, many hateful tweets have been mislabelled. It is advisable to look at the outcomes and faults in much detail, in order to comprehend the misclassification. Our analysis shows a better outcome in detecting hateful speech in social media.

From fear to hate: Sources of anti‐Asian sentiment during COVID‐19

AbstractObjectiveThis paper seeks to explain and empirically test how a public health crisis, such as the COVID‐19 pandemic, might lead to discriminatory attitudes and behaviors against marginalized groups.MethodWe identify four causal mechanisms that may account for the increase in anti‐Asian racism during COVID‐19. Using a large data set of geolocated COVID‐19‐related tweets in the U.S., we examine the spatiotemporal pattern of anti‐Asian sentiment on Twitter and test all four mechanisms that link the crisis to the behaviors of “othering” and “scapegoating.”ResultsWe find evidence consistent with the elite cueing, perceived threat, and grievance mechanisms. The president's scapegoating rhetoric significantly increased anti‐Asian hate tweets across all counties. In addition, high COVID‐19 infection rates and grievances from containment policies are also associated with greater numbers of anti‐Asian hate tweets. On the other hand, counties with larger vulnerable populations do not seem to have more anti‐Asian hate tweets.ConclusionsWe demonstrate that a crisis alone does not necessarily lead to othering behavior, while elite rhetoric can significantly influence the public's opinion and behavior during a crisis.

Violence against women on Twitter in India: Testing a taxonomy for online misogyny and measuring its prevalence during COVID-19.

Online misogyny is a violation of women's digital rights. Empirical studies on this topic are however lacking, particularly in low- and middle- income countries. The current study aimed to estimate whether prevalence of online misogyny on Twitter in India changed since the pandemic. Based on prior theoretical work, we defined online misogyny as consisting of six overlapping forms: sexist abuses, sexual objectification, threatening to physically or sexually harm women, asserting women's inferiority, justifying violence against women, and dismissing feminist efforts. Qualitative analysis of a small subset of tweets posted from India (40,672 tweets) substantiated this definition and taxonomy for online misogyny. Supervised machine learning models were used to predict the status of misogyny across a corpus of 30 million tweets posted from India between 2018 and 2021. Next, interrupted time series analysis examined changes in online misogyny prevalence, before and during COVID-19. Qualitative assessment showed that online misogyny in India existed most in the form of sexual objectification and sexist abusive content, which demeans women and shames them for their presumed sexual activity. Around 2% of overall tweets posted from India between 2018 and 2021 included some form of misogynistic content. The absolute volume as well as proportion of misogynistic tweets showed significant increasing trends after the onset of COVID-19, relative to trends prior to the pandemic. Findings highlight increasing gender inequalities on Twitter since the pandemic. Aggressive and hateful tweets that target women attempt to reinforce traditional gender norms, especially those relating to idealized sexual behavior and framing of women as sexual beings. There is an urgent need for future research and development of interventions to make digital spaces gender equitable and welcoming to women.

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.

Emojis as anchors to detect Arabic offensive language and hate speech

AbstractWe introduce a generic, language-independent method to collect a large percentage of offensive and hate tweets regardless of their topics or genres. We harness the extralinguistic information embedded in the emojis to collect a large number of offensive tweets. We apply the proposed method on Arabic tweets and compare it with English tweets—analyzing key cultural differences. We observed a constant usage of these emojis to represent offensiveness throughout different timespans on Twitter. We manually annotate and publicly release the largest Arabic dataset for offensive, fine-grained hate speech, vulgar, and violence content. Furthermore, we benchmark the dataset for detecting offensiveness and hate speech using different transformer architectures and perform in-depth linguistic analysis. We evaluate our models on external datasets—a Twitter dataset collected using a completely different method, and a multi-platform dataset containing comments from Twitter, YouTube, and Facebook, for assessing generalization capability. Competitive results on these datasets suggest that the data collected using our method capture universal characteristics of offensive language. Our findings also highlight the common words used in offensive communications, common targets for hate speech, specific patterns in violence tweets, and pinpoint common classification errors that can be attributed to limitations of NLP models. We observe that even state-of-the-art transformer models may fail to take into account culture, background, and context or understand nuances present in real-world data such as sarcasm.

The Effects of Observer Expectations on Judgments of Anti-Asian Hate Tweets and Online Activism Response

The rise of racial hate speech on social media has raised critical questions for scholars to explore. It is necessary to understand how outside observers passively evaluate (a) online racial hate speech posts on social media and (b) whether those evaluations are related to observers’ subsequent behavior. This study explored how observers evaluate acts of majority-on-minority and minority-on-minority anti-Asian hate tweets on Twitter. In an experiment ( n = 196) informed by expectancy violations theory, we tested how White observers evaluated anti-Asian tweets ostensibly posted by either a White or Black source. Analysis revealed a moderated-mediation pathway in which observers’ political partisanship (Democrat/Republican) affected how they judged the ethnic prototypicality of White and Black sources of racial hate speech; these source prototypicality judgments were in turn associated with observers’ judgments of tweet offensiveness and self-reported intentions to engage in online activism (i.e., signing an online petition). These results contribute to our understanding of outside observers’ differential expectancies regarding online hate speech, and how those expectancies can affect perceptions of and reactions to acts of racism.

Social media content classification and community detection using deep learning and graph analytics

Social cybermedia has greatly changed the world’s perspective on information sharing and propagation. These platforms build virtual user communities in the form of followings or followers that can spread positive or negative content. These virtual communities play a vital role in the spread of hate content, which may create a severe law-and-order situation and social unrest in any country. Hate content can include offensive, anti-state, anti-religion, sexist, and racist posts. Classifying such content and identifying influential individuals who participate in these communities are challenging tasks for law enforcement agencies. This study proposes novel deep learning and graph-based approaches to identifying hate content, followed by approaches to detecting communities and exploring social media to detect hate content. Twitter is used as a case study, and tweets are extracted and annotated by linguistic experts to develop a dataset for experimentation and validation. The proposed customized LSTM-GRU model is used to classify hate content into six categories. The developed model offers an accuracy of 98.14% on the obtained dataset. The Girvan–Newman algorithm is employed for Twitter community detection and successfully identifies the most influential individuals. Moreover, the proposed approach precisely detects intraclass communities and highly influential persons. The developed model effectively identifies hate tweets and communities and can be used to monitor social media to identify any potential threat of unrest.

An Analytical Study of Translating Hate Speech Tweets from English into Arabic

Social media, nowadays, is considered a mirror in which all human beings can see their true identity, a sensor to measure used to see the pulse of society. It is a weapon used by individual and politicians alike to attack each other. In this paper, the researcher aims to investigate and analysis the translation of ten hateful tweets by students in the second stage. The students were asked to implement Vinay and Darbelnet (1995)’ procedures of translation. The results of this investigation shed some lights on the problems the students encounter, and over solutions to be followed. Keywords: Social Media, Twitter, Hateful Tweets, discrimination, Translation Procedure.

Analyzing social media for measuring public attitudes toward controversies and their driving factors: a case study of migration

Among other ways of expressing opinions on media such as blogs, and forums, social media (such as Twitter) has become one of the most widely used channels by populations for expressing their opinions. With an increasing interest in the topic of migration in Europe, it is important to process and analyze these opinions. To this end, this study aims at measuring the public attitudes toward migration in terms of sentiments and hate speech from a large number of tweets crawled on the decisive topic of migration. This study introduces a knowledge base (KB) of anonymized migration-related annotated tweets termed as MigrationsKB (MGKB). The tweets from 2013 to July 2021 in the European countries that are hosts of immigrants are collected, pre-processed, and filtered using advanced topic modeling techniques. BERT-based entity linking and sentiment analysis, complemented by attention-based hate speech detection, are performed to annotate the curated tweets. Moreover, external databases are used to identify the potential social and economic factors causing negative public attitudes toward migration. The analysis aligns with the hypothesis that the countries with more migrants have fewer negative and hateful tweets. To further promote research in the interdisciplinary fields of social sciences and computer science, the outcomes are integrated into MGKB, which significantly extends the existing ontology to consider the public attitudes toward migrations and economic indicators. This study further discusses the use-cases and exploitation of MGKB. Finally, MGKB is made publicly available, fully supporting the FAIR principles.

Temperature impacts on hate speech online: evidence from 4 billion geolocated tweets from the USA

A link between weather and aggression in the offline world has been established across a variety of societal settings. Simultaneously, the rapid digitalisation of nearly every aspect of everyday life has led to a high frequency of interpersonal conflicts online. Hate speech online has become a prevalent problem that has been shown to aggravate mental health conditions, especially among young people and marginalised groups. We examine the effect of temperature on the occurrence of hate speech on the social media platform Twitter and interpret the results in the context of the interlinkage between climate change, human behaviour, and mental health. In this quantitative empirical study, we used a supervised machine learning approach to identify hate speech in a dataset containing around 4 billion geolocated tweets from 773 cities across the USA between May 1, 2014 and May 1, 2020. We statistically evaluated the changes in daily hate tweets against changes in local temperature, isolating the temperature influence from confounding factors using binned panel-regression models. The prevalence of hate tweets was lowest at moderate temperatures (12 to 21°C) and marked increases in the number of hate tweets were observed at hotter and colder temperatures, reaching up to 12·5% (95% CI 8·0-16·5) for cold temperature extremes (-6 to -3°C) and up to 22·0% (95% CI 20·5-23·5) for hot temperature extremes (42 to 45°C). Outside of the moderate temperature range, the hate tweets also increased as a proportion of total tweeting activity. The quasi-quadratic shape of the temperature-hate tweet curve was robust across varying climate zones, income quartiles, religious and political beliefs, and both city-level and state-level aggregations. However, temperature ranges with the lowest prevalence of hate tweets were centred around the local temperature mean and the magnitude of the increases in hate tweets for hot and cold temperatures varied across the climate zones. Our results highlight hate speech online as a potential channel through which temperature alters interpersonal conflict and societal aggression. We provide empirical evidence that hot and cold temperatures can aggravate aggressive tendencies online. The prevalence of the results across climatic and socioeconomic subgroups points to limitations in the ability of humans to adapt to temperature extremes. Volkswagen Foundation.

Pinpointing Fine-Grained Relationships between Hateful Tweets and Replies

Recent studies in the hate and counter hate domain have provided the grounds for investigating how to detect this pervasive content in social media. These studies mostly work with synthetic replies to hateful content written by annotators on demand rather than replies written by real users. We argue that working with naturally occurring replies to hateful content is key to study the problem. Building on this motivation, we create a corpus of 5,652 hateful tweets and replies. We analyze their fine-grained relationships by indicating whether the reply (a) is hate or counter hate speech, (b) provides a justification, (c) attacks the author of the tweet, and (d) adds additional hate. We also present linguistic insights into the language people use depending on these fine-grained relationships. Experimental results show improvements (a) taking into account the hateful tweet in addition to the reply and (b) pretraining with related tasks.

Spatial Distribution of Hateful Tweets Against Asians and Asian Americans During the COVID-19 Pandemic, November 2019 to May 2020.

Objectives. To illustrate the spatiotemporal distribution of geolocated tweets that contain anti-Asian hate language in the contiguous United States during the early phase of the COVID-19 pandemic. Methods. We used a data set of geolocated tweets that match with keywords reflecting COVID-19 and anti-Asian hate and identified geographical clusters using the space-time scan statistic with Bernoulli model. Results. Anti-Asian hate language surged between January and March 2020. We found clusters of hate across the contiguous United States. The strongest cluster consisted of a single county (Ross County, Ohio), where the proportion of hateful tweets was 312.13 times higher than for the rest of the country. Conclusions. Anti-Asian hate on Twitter exhibits a significantly clustered spatiotemporal distribution. Clusters vary in size, duration, strength, and location and are scattered across the entire contiguous United States. Public Health Implications. Our results can inform decision-makers in public health and safety for allocating resources for place-based preparedness and response for pandemic-induced racism as a public health threat. (Am J Public Health. 2022;112(4):646-649. https://doi.org/10.2105/AJPH.2021.306653.

Detecting Islamic Radicalism Arabic Tweets Using Natural Language Processing

The image of the tolerant religion of Islam has been distorted by extremists in the last two decades in many ways, such as luring teenagers into terrorist acts. Nowadays, millions of users socialize and share ideas using social media platforms such as Twitter. Typically, the ideas shared on Twitter (tweets) reach and influence many people who could simply retweet them and make them even spread faster. Unfortunately, some of these ideas are posted by extremists who share hateful Arabic content. Thus, it is very important to automate the process of controlling and monitoring hateful Arabic tweets, given that Arabic is the most widely used language in the Islamic world. In this paper, we provide a manually labeled and curated dataset of 3,000 Arabic tweets that contain hateful and non-hateful tweets. To automate the process of detecting hateful tweets, we utilize advanced Machine Learning (ML) techniques and perform sentiment analysis to capture the meaning of the Arabic words in a proper word embedding. Also, we used the proposed model to classify and analyze 100,000 tweets of the last decade. The outcome of this work promotes future research on analyzing Arabic hateful speech by providing a manually labeled Arabic dataset, and the trained model (achieved 92% accuracy) which can be used as an underlying tool by governments, Internet service providers, and social media applications to detect any inflammatory tweets before they spread to a wider audience.

The Not-So-Obvious Harm of Cyberhate: Source Magnification of Hate Tweets, Unhealthy Food Choice, and the Moderating Role of Group Identification

Guided by research demonstrating the impact of social identity threat on physical health and the source magnification framework, we investigated how hate tweets from multiple sources, relative to hate tweets from a single source, influence target group members’ health-related behavior-specifically, food choice. We conducted an online experiment with a sample of Asian Twitter users in the United States. Participants were randomly assigned to view an identical set of anti-Asian hate tweets either from a single individual or from multiple individuals. Subsequently, in an ostensibly unrelated task masked as a marketing survey, participants customized a pizza by selecting toppings from a list presenting an equal number of healthy and unhealthy toppings. We found that participants exposed to multiple-source hate tweets selected a greater number of unhealthy toppings than those exposed to single-source hate tweets. We also found that group identification was a significant moderator, with those lower in group identification more likely to choose unhealthy toppings when exposed to multiple-source hate tweets. Our findings suggest that exposure to multiple-source hate messages, relative to single-source hate messages, may exert a detrimental influence on target group members’ health to a greater extent, while highlighting a possible buffering role of group identification.

COVID-19: A Pandemic of Anti-Asian Cyberhate

Hateful online attacks targeting individuals of Asian descent have increased dramatically in the wake of the coronavirus pandemic. This study examines daily trends in inflammatory keywords associated with anti-Asian cyberhate using data collected from Twitter between January 1, 2020 and February 28, 2021. Additionally, we explore the sentiment of tweets and track the evolution of discourse on Twitter regarding Asians and the coronavirus. Our results indicate that the usage of inflammatory keywords associated with anti-Asian hate on Twitter rose dramatically at the outset of the outbreak, and then spiked at various points over the next fourteen months. Notably, offline events and rhetoric appear to have affected the rate of hateful tweets about the pandemic targeting Asians. Additionally, we find that discourse on Twitter around the words China, Chinese, Asians, and virus has become increasingly negative and stigmatizing during the course of the coronavirus outbreak.

The Effects of Racial Hate Tweets on Perceived Political Polarization and the Roles of Negative Emotions and Individuation

While there are growing concerns among Americans about widening partisan gaps, some claim that the perceptions of polarization are exaggerated. According to the theory of false polarization, people tend to assume that opposing parties are more ideologically extreme than they actually are. Hate speech expressed and disseminated through social media, which provokes anger among the audience, might be a factor that drives such heightened perceptions. Drawing on the social identity model of deindividuation effects (SIDE model) and the affect-cognition literature, this experimental study (N = 480) investigates whether exposure to hate messages on Twitter (i.e., hate tweets) targeting Whites triggers negative emotions (i.e., anger and fear), thereby increasing the perceptions of political polarization among Whites. This study also examines whether the effect of hate tweets decreases when the source is individuated with personal information disclosed on the profile. A conditional process analysis reveals that exposure to hate tweets significantly provokes negative emotions and heightens perceived polarization through anger, and that source individuation significantly moderates the anger-provoking effect of hate tweets. These findings indicate that online hate speech may have implications for political perception and anger can play a significant role in the process. The theoretical and practical implications are discussed.

Where do angry birds tweet? Income inequality and online hate in Italy

AbstractDo spatial socioeconomic features influence a digital behaviour like cyberhate? Our contribution provides an answer to this question, showing how high levels of income inequality determine high volumes of hate tweets in Italy. Our findings are robust to potential endogeneity problems of income inequality, as well as to the inclusion of confounding factors and to competing estimation strategies. Additionally, we find that education does not act as a protective factor against cyberhate in unequal places, aligning with existing evidence showing that inequality may trigger intolerance, including among educated people, threatening the perceived stability of social positions. Also, in the Italian case, the perception of economic insecurity fuels cyberhate, alongside the transmission of self-interest values along family generations. The latter finding relates to existing evidence supporting the role of persistent social norms in shaping people’s attitudes.

A probabilistic clustering model for hate speech classification in twitter

The key challenges for automatic hate-speech classification in Twitter are the lack of generic architecture, imprecision, threshold settings and fragmentation issues. Most studies used binary classifiers for hate speech classification, but these classifiers cannot really capture other emotions that may overlap between positive or negative class. Hence, a probabilistic clustering model for hate speech classification in twitter was developed to tackle problems with hate speech classification. A metadata extractor was used to collect tweets containing hate speech keywords and a crowd-sourced experts was employed to label the collected hate tweets into two categories: hate speech and non-hate speech. Features representation was done with Term Frequency- Inverse Document Frequency (TF-IDF) model and enhanced with topics inferred by a Bayes classifier. A rule-based clustering method was used to automatically classify real-time tweets into the correct topic clusters. Fuzzy logic was then used for hate speech classification using semantic fuzzy rules and a score computation module. From the evaluation results, it was observed that the developed model performed better in hate speech detection with F1-sore of 0.9256 using a 5-fold cross validation. Similarly, the developed model for hate speech classification performed better with F1-score of 91.5 compared to related models. The developed model also indicates a more perfect test having an AUC of 0.9645, when compared to similar methods. The Paired Sample t-Test validated the efficiency of the developed model for hate speech classification.

Detection of hate speech in Arabic tweets using deep learning

Nowadays, people are communicating through social networks everywhere. However, for whatever reason it is noticeable that verbal misbehaviors, such as hate speech is now propagated through the social networks. One of the most popular social networks is Twitter which has gained widespread in the Arabic region. This research aims to identify and classify Arabic tweets into 5 distinct classes: none, religious, racial, sexism or general hate. A dataset of 11 K tweets was collected and labelled and SVM model was used as a baseline to be compared against 4 deep learning models: LTSM, CNN + LTSM, GRU and CNN + GRU. The results show that all the 4 deep learning models outperform the SVM model in detecting hateful tweets. Although the SVM achieves an overall recall of 74%, the deep learning models have an average recall of 75%. However, adding a layer of CNN to LTSM enhances the overall performance of detection with 72% precision, 75% recall and 73% F1 score.

Classifying COVID-19-related hate Twitter users using deep neural networks with sentiment-based features and geopolitical factors

Anti-Asian hate tweets caused by COVID-19 pandemic is an ongoing social problem in the USA and around the world. Although existing studies have been done by using a text classifier, little is known on how deep learning works with public sentiments of political opinions and geographical diversities. This paper provides a new method to classify the pandemic-related anti-Asian hater on Twitter. A novel dataset for tracking pandemic-related Twitter users, which contains more than 10 million tweets, is created in this study. Target users are annotated by identifying their sentiments towards the US elections with their geolocations. The empirical result indicates that the political sentiments and the county-level election results make significant contributions to the model building. By training a DNN model, over 190,000 Twitter users are classified as hate or non-hate with a 61% accuracy and a 0.63 AUC score.