The increasing prevalence of content on social media and online information consumption has accelerated the spread of disinformation, posing a significant threat to societal stability and security. Recent events — from geopolitical conflicts to public health crises — demonstrate how disinformation can influence public opinion. In response, we take a systematic, scientifically grounded approach to understanding and countering disinformation narratives.
The urgent need to address this pervasive influence of disinformation, especially in light of rapidly changing digital media and ever-growing quantities of AI-generated content informs our central research question:
How can disinformation be concretely represented through narratives to provide a comprehensive overview of past, current, and emerging trends in disinformation?
By focusing on narrative structures, we aim to create a robust framework for analyzing disinformation in a clear and transparent manner. VeraXtract is structured around three overarching topics:
Narrative Extraction: We develop methods to extract narratives from diverse sources and modalities — including media outlets, blogs, social media, and messaging channels. This involves defining what constitutes a narrative in the context of disinformation and applying techniques such as topic modeling, event extraction, summarization, and entity recognition.
Knowledge Base/Graph Development: A dynamic knowledge base is being created to aggregate verified facts and documented disinformation. This resource will allow the visualization and analysis of trends by linking information across different time points and levels of detail in a knowledge graph.
Explainable AI: Recognizing the importance of transparency, we place a strong emphasis on explainability in our systems. We develop methods that generate clear, understandable explanations for AI-driven analyses and evaluate them empirically, thereby fostering trust and facilitating informed decision-making among users.
These main objectives are realized on three interconnected levels of abstraction:
Level 1 – Narrative Representation: Extraction of narratives from continuously monitored text and image sources, including contributions from accredited fact-checkers and social media platforms.
Level 2 – Thematic Analysis: A deeper examination of the underlying topics within these narratives, enabling users to track, understand, and analyze the evolution of disinformation themes.
Level 3 – Entity Analysis: Investigation of key entities such as individuals, organizations, and locations. This level also involves identifying disinformation superspreaders and understanding their roles in disseminating false information.
Integrating these research efforts delivers a comprehensive framework for understanding and combating disinformation in Germany. Our approach emphasizes transparency, traceability, and practical applicability, ensuring that our findings can inform both academic research and policy-making in a realistic and grounded manner.
Verification and Extraction of Disinformation Narratives with Individualized Explanations