Projects
Image contextualization for fact-checking
🗒️Summary
Most research in multimodal fact-checking focuses on predicting the veracity of multimodal claims. However, in real-world fact-checking practices, particular attention is also given to identifying the true context of misrepresented or manipulated images and videos. For example, identifying the true date, location, or depicted event of an image used out of its original context.
In this project, we assemble datasets and propose methods to assist human fact-checkers in predicting the true context of multimodal misinformation content.
📖 Related works
Countering and detecting misleading charts
🗒️Summary
Charts are a convenient way to communicate data insights. However, they can also be used to misinform readers by distorting the underlying data, for example, by truncating or inverting the axes.
In this project, we evaluate whether MLLMs are vulnerable to such misleading visualizations and we propose detection and correction methods to mitigate their negative effects.
Check out our “Awesome misleading visualizations” repo to learn more about the relevant resources, papers, and datasets.
📖 Related works
- Protecting MLLMs against misleading visualizations (ACL 2026)
- Misviz and Misviz-synth (ACL 2026)
- ChartAttack (EMNLP 2026)