Identifying Misinformation Spreaders through Multi-Source User Modeling

Social media platforms have become a major channel for information consumption, but also facilitate the rapid spread of misinformation. While much prior work focuses on detecting misleading content, identifying users who are likely to spread misinformation remains a more challenging task, as such behavior depends not only on textual signals but also on interaction patterns and temporal dynamics. In this paper, we model misinformation spreader detection as a user-level behavior prediction problem and propose a hybrid architecture that combines content-derived user features with graph-based representations of user similarity, user interactions, and content interactions. Experimental results on a benchmark dataset show that integrating multiple user representations improves misinformation spreader detection. These findings highlight the potential of multi-source user representations for user-centered misinformation detection.