A hybrid AI Approach for Understanding Science in Online Discourse

Leader: Prof. Dr. Stefan Dietze
Scientific unit: Angewandte Informatik


Scientific findings form a central part of public discourse on the web and in social media, e.g., in the context of the COVID-19 pandemic. However, due to the inherent complexity of scientific statements, as well as the inherent mechanisms of AI-driven online platforms where controversial or false statements have been shown to generate more interactions and interest, scientific evidence is often presented in simplified, decontextualized, and misleading ways. Recent examples include statements about vaccine risks or COVID19 mortality, which have often been presented in truncated or misleading ways on the web.

AI4Sci addresses the challenge of developing hybrid artificial intelligence (AI) methods for detecting and interpreting scientific claims in large datasets from online discourse, and thus countering disinformation in the context of science communication. To this end, the project builds on advances in areas such as deep learning, natural language processing, and knowledge graphs, and will develop methods to detect, for example, the quality, correctness, or completeness of scientific claims in social media or on news sites by establishing and using relationships between claims and their primary scientific sources. 

This creates tools for tracking scientific findings in online discourse and detecting misinformation to improve public discourse and understanding of complex scientific issues, thereby exerting a democracy-enhancing influence on online communication. AI4Sci's hybrid methodology will also contribute to important challenges in AI, such as transparency and reproducibility of AI models. As part of an international collaboration between GESIS (Germany) and LIRMM (France) and with the involvement of regional institutions such as the Heine Center for Artifical Intelligence and Data Science (HeiCAD) at HHU Düsseldorf, a regional, internationally networked hub for AI methods for the use and analysis of online discourse will be established, thus strengthening Germany as an international hub for AI research.



Sponsored by



LIRRM - Labor für Informatik, Robotik und Mikroelektronik von Montpellier, Universität Montpellier, Frankreich


Boland, K., Fafalios, P., Tchechmedjiev, A., Dietze, S., Todorov, K., Beyond Facts – a Survey and Conceptualisation of Claims in Online Discourse Analysis, Semantic Web Journal, IOS Press 2021.

Zhu, X., Zhu, L., Guo, J., Liang, S., Dietze, S., GL-GCN: Global and Local Dependency Guided Graph Convolutional Networks for aspect-based sentiment classification, Expert Systems with Applications, Volume 186, 115712, Elsevier, 2021.

Dimitrov, D., Baran, E., Fafalios, P., Yu, R., Zhu, X., Zloch, M., Dietze, S., TweetsCOV19 – A Knowledge Base of Semantically Annotated Tweets about the COVID-19 Pandemic, 29th ACM International Conference on Information & Knowledge Management (CIKM2020), ACM 2020