Digital Library
Decoding Antisemitism An AI-driven Study on Hate Speech & Imagery Online First Discourse Report
Topic:
Antisemitism & Antizionism, Israel & Regional Politics
Principal Investigators:
Dr. Matthias J. Becker, Hagen Troschke, Dr. Daniel Allington
Study Date:
2021
Source:
Alfred Landecker Foundation,Center for Research on Antisemitism,King's College London (KCL),Technische Universität Berlin
Key Findings:
The interdisciplinary project titled "Decoding Antisemitism" is centered on investigating antisemitism on the internet, particularly within the comments sections of media and social platforms in Germany, France, and Britain. Based at the Center for Research on Antisemitism at TU Berlin, the project spans three years and collaborates with King's College London.
Employing a mixed methods approach, including qualitative, quantitative, and AI-aided analyses, the project aims to comprehensively examine various forms of antisemitism discourse. Notably, it focuses on politically moderate environments, filling a research gap and providing insights into the extent and nature of online antisemitism. The project involves a diverse advisory board and seeks to bridge humanities with AI and web-related disciplines. It emphasizes the importance of understanding and addressing online hate speech, considering its societal implications and potential for real-world violence.
The project's goals include cataloging antisemitic hate speech, emphasizing implicit forms often overlooked in previous studies, and producing half-yearly discourse reports on current trends. Recognizing the increasing prevalence of online antisemitism and its societal impacts, the project aims to provide actionable insights for intervention and prevention efforts. It plans to develop an open-source tool for platform moderators and establish an internet institute to facilitate knowledge transfer and expand research across Europe. At the point of publishing, the project was in its initial stages of data collection and analysis, with plans to submit the first comprehensive report by summer 2021.
Methodology:
A three-step process to analyze antisemitic discourse on online platforms is employed. Firstly, user comments from mainstream media websites and Facebook pages in Germany, Britain, and France are collected and stored in anonymized or pseudonymized forms. Reference corpora are also created to contrast with the main corpus, allowing for comparisons to determine the presence of antisemitism based on specific discourse triggers.
The study design begins with qualitative content analysis to identify antisemitic patterns of language use. Categories are developed inductively to capture both explicit and implicit expressions of antisemitism. Machine learning techniques are then applied to categorize comments accurately, using the qualitative results as training data to distinguish between antisemitic and non-antisemitic comments. Continuous updates to the algorithm allow for adaptation to evolving antisemitic discourse.
Then, quantitative analyses (focusing on frequency, collocation, and vector analyses) are conducted based on the qualitative findings. This multi-step approach is chosen due to the implicit nature of antisemitic expressions in mainstream discourse, which may not be captured effectively through initial quantitative analyses alone.
A code system is developed using MAXQDA to map the diversity of antisemitic expressions, including stereotypes, conspiracy theories, and linguistic patterns. The system incorporates both conceptual and linguistic levels of analysis, ensuring comprehensive coverage of antisemitic discourse. Intercoder reliability is ensured through consensus validation, and semiotic and visual elements are also considered, reflecting the significance of non-verbal communication in online discourse. Ultimately, this methodological approach aims to provide reliable insights into the nature and frequency of online antisemitism across different language communities.