Debiased Social Media Fake News Detection based on User Portraits Simulation

ACL ARR 2025 February Submission7308 Authors

Abstract: Social media fake news detection aims to detect fake news from platforms through online interaction data, which mainly consists of user posts and related comments. Through statistics, we found that the number of replies to posts depends largely on the time of posting, which we named temporal bias of data. Traditional methods focus on graph modeling to explore the potential structures among social texts, but ignore data bias. Although related methods based on large language models (LLMs) generate interactive comments and perform input enhancement, the generated information is uncontrollable and does not address data bias. In response, we propose a approach that uses LLMs to debias through data augmentation, named DUPS. The method first uses the LLM to analyze the user portraits, and then simulates the corresponding portrait to generate interactive comments, thereby reconstructing unbiased data. Experimental results on three datasets show that DUPS outperforms the current State-Of-The-Art approaches.

Paper Type: Long

Research Area: NLP Applications

Research Area Keywords: fact checking, rumor/misinformation detection

Languages Studied: English

Submission Number: 7308