Online Library

Research examining radical visual communication and its manifestation on the trending platform TikTok is limited. This paper presents a novel methodological framework for studying mobilization strategies of far-right groups on TikTok, employing a mixed-method approach that combines manual annotation, unsupervised image classification, and named-entity recognition to analyze the dynamics of radical visuals at scale. Differentiating between internal and external mobilization, we use popularity and engagement cues to investigate far-right mobilization efforts on TikTok within and outside their community. Our findings shed light on the effectiveness of unsupervised image classification when utilized within a broader mixed-method framework, as each observed far-right group employs unique platform characteristics. While Conspiracists flourish in terms of overall popularity and internal mobilization, nationalist and protest content succeeds by using a variety of persuasive visual content to attract and engage external audiences. The study contributes to existing literature by bridging the gap between visual political communication at scale and radicalization research. By offering insights into mobilization strategies of far-right groups, our study provides a foundation for policymakers, researchers, and online platforms to develop proactive measures to address the risks associated with the dissemination of extremist ideologies on social media.

Do social media companies exacerbate antisemitism and hate through their own recommendation and amplification tools? We investigated how four of the biggest social media platforms treated users who searched for or engaged with content related to anti-Jewish tropes, conspiracy theories, and other topics. Three of them–Facebook, Instagram, and Twitter1 (now known as X after a rebranding by owner Elon Musk)–filled their feeds with far more antisemitic content and conspiratorial accounts and recommended the most hateful influencers and pages to follow. One platform, YouTube, did not.

In this joint study by the ADL Center for Technology and Society (CTS) and Tech Transparency Project (TTP), researchers created six test personas of different ages and genders and set up accounts for them on Facebook, Instagram, Twitter, and YouTube. Researchers had them search each platform for a basket of terms related to conspiracy theories as well as popular internet personalities, commentators, and video games. The study then examined the content that the platforms’ algorithms recommended to these hypothetical users.

Do social media and search companies exacerbate antisemitism and hate through their own design and system functions? In this joint study by the ADL Center for Technology and Society (CTS) and Tech Transparency Project (TTP), we investigated search functions on both social media platforms and Google. Our results show how these companies’ own tools–such as autocomplete and auto-generation of content–made finding and engaging with antisemitism easier and faster.1 In some cases, the companies even helped create the content themselves.

Our researchers compiled a list of 130 hate groups and movements from ADL’s Glossary of Extremism, picking terms that were tagged in the glossary with all three of the following categories: “groups/movements,” “white supremacist,” and “antisemitism.”2 The researchers then typed each term into the respective search bars of Facebook, Instagram and YouTube, and recorded the results.

Social media platforms play a key role in fostering the outreach of extremism by influencing the views, opinions, and perceptions of people. These platforms are increasingly exploited by extremist elements for spreading propaganda, radicalizing, and recruiting youth. Hence, research on extremism detection on social media platforms is essential to curb its influence and ill effects. A study of existing literature on extremism detection reveals that it is restricted to a specific ideology, binary classification with limited insights on extremism text, and manual data validation methods to check data quality. In existing research studies, researchers have used datasets limited to a single ideology. As a result, they face serious issues such as class imbalance, limited insights with class labels, and a lack of automated data validation methods. A major contribution of this work is a balanced extremism text dataset, versatile with multiple ideologies verified by robust data validation methods for classifying extremism text into popular extremism types such as propaganda, radicalization, and recruitment. The presented extremism text dataset is a generalization of multiple ideologies such as the standard ISIS dataset, GAB White Supremacist dataset, and recent Twitter tweets on ISIS and white supremacist ideology. The dataset is analyzed to extract features for the three focused classes in extremism with TF‐IDF unigram, bigrams, and trigrams features. Additionally, pretrained word2vec features are used for semantic analysis. The extracted features in the proposed dataset are evaluated using machine learning classification algorithms such as multinomial Naïve Bayes, support vector machine, random forest, and XGBoost algorithms. The best results were achieved by support vector machine using the TF‐IDF unigram model confirming 0.67 F1 score. The proposed multi‐ideology and multiclass dataset shows comparable performance to the existing datasets limited to single ideology and binary labels.

The whitepaper has three parts. The first part provides some contextual background, describing the diverse range of online services utilised by terrorists and extremists and the process by which propaganda is disseminated online. The second part details industry responses. As well as referrals from users and law enforcement, it describes the use of AI for proactive detection and collaborative, cross-platform initiatives. The third part describes four issues for discussion: transparency; definitional clarity; the impact on those targeted; and, the use of online data for predictive purposes.