New Fake News Detection Tackles the 'Cold User' Problem

Most fake news detection models today rely on signals like past user behavior and engagement history. But what happens when a user is brand new or has barely interacted? These 'cold users'—a term borrowed from recommender systems—break existing approaches. A team of researchers from India (Karnam, Kundu, Arora, Jain, Mukherjee) identified that cold users are actually widespread in real-world social media datasets, yet largely ignored by current state-of-the-art detectors. Their paper, accepted at ICWSM 2026, proposes a solution: USER EVIDENCE NETWORK (UEN). This socially-aware context representation doesn't require past behavior data for cold users. Instead, it approximates missing signals by leveraging interactions from similar existing users and the network structure. The result is a robust method that can flag misinformation even when the poster has zero prior footprint.

This matters because platforms like Twitter and Facebook constantly face new accounts spreading viral falsehoods. By addressing the cold user challenge, UEN promises more equitable and effective detection—without needing to wait for a user to build a history. The paper also establishes the prevalence of cold users through real-world data analysis, making the case that ignoring them leaves a major blind spot in online safety systems. For professionals building trust and safety infrastructure, this work points to a critical missing piece in current moderation pipelines.