LLM-Generated Fake News Induces Truth Decay in News Ecosystem: A Case Study on Neural News Recommendation

2025-04-28

AI Summary

This paper identifies the “truth decay” phenomenon—where authentic news items are systematically demoted in neural news recommendation systems due to competition with large language model (LLM)-generated fake news. We construct a scalable simulation pipeline containing 56K LLM-generated news articles, enabling the first formal definition and empirical validation of truth decay. Drawing on user familiarity as a key behavioral signal, we explain how fake news gains ranking advantage and uncover a significant positive correlation between news perplexity and recommendation rank. Our contributions are threefold: (1) the first quantitative formulation and measurement of truth decay rate; (2) a controllable fake news dataset and benchmark evaluation framework for truth decay analysis; and (3) a novel defense method integrating credibility-aware source enhancement and perplexity-sensitive re-ranking, which effectively mitigates truth decay while preserving recommendation accuracy.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Learning Preferences or Rankings
• Application Domains: Misinformation & Fake News

Application Category

• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity

Abstract

Online fake news moderation now faces a new challenge brought by the malicious use of large language models (LLMs) in fake news production. Though existing works have shown LLM-generated fake news is hard to detect from an individual aspect, it remains underexplored how its large-scale release will impact the news ecosystem. In this study, we develop a simulation pipeline and a dataset with ~56k generated news of diverse types to investigate the effects of LLM-generated fake news within neural news recommendation systems. Our findings expose a truth decay phenomenon, where real news is gradually losing its advantageous position in news ranking against fake news as LLM-generated news is involved in news recommendation. We further provide an explanation about why truth decay occurs from a familiarity perspective and show the positive correlation between perplexity and news ranking. Finally, we discuss the threats of LLM-generated fake news and provide possible countermeasures. We urge stakeholders to address this emerging challenge to preserve the integrity of news ecosystems.

Problem

• Investigating impact of LLM-generated fake news on news ecosystems
• Analyzing truth decay in neural news recommendation systems
• Exploring detection and countermeasures for LLM-produced fake news

Innovation

• Simulation pipeline for LLM-generated fake news impact
• Dataset with 56k diverse fake news samples
• Analysis of truth decay in neural recommendation systems