Incentive-Aligned Multi-Source LLM Summaries

2025-09-29

AI Summary

Current large language models struggle with factual accuracy in multi-source summarization, are vulnerable to adversarial content, and lack effective incentives for source honesty. This paper proposes the Truthful Text Summarization (TTS) framework: first, decomposing synthetic texts into atomic propositions and extracting source stances; second, assessing source consistency via an improved multi-task peer-prediction mechanism to dynamically filter unreliable sources; and third, regenerating the summary. TTS is the first framework to achieve source incentive alignment without ground-truth labels—ensuring honest reporting constitutes a Nash equilibrium strategy. Experiments demonstrate that TTS significantly improves factual accuracy and adversarial robustness while preserving linguistic fluency, thereby enabling positive synergy between information credibility and visibility.

Technology Category

• Natural Language Processing: Summarization
• Machine Learning: Large Multimodal Models (LMMs)
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Web data provenance, reliability, and authenticity

Abstract

Large language models (LLMs) are increasingly used in modern search and answer systems to synthesize multiple, sometimes conflicting, texts into a single response, yet current pipelines offer weak incentives for sources to be accurate and are vulnerable to adversarial content. We introduce Truthful Text Summarization (TTS), an incentive-aligned framework that improves factual robustness without ground-truth labels. TTS (i) decomposes a draft synthesis into atomic claims, (ii) elicits each source's stance on every claim, (iii) scores sources with an adapted multi-task peer-prediction mechanism that rewards informative agreement, and (iv) filters unreliable sources before re-summarizing. We establish formal guarantees that align a source's incentives with informative honesty, making truthful reporting the utility-maximizing strategy. Experiments show that TTS improves factual accuracy and robustness while preserving fluency, aligning exposure with informative corroboration and disincentivizing manipulation.

Problem

• Improving factual robustness in LLM summaries without ground-truth labels
• Aligning source incentives with informative honesty to maximize utility
• Filtering unreliable sources before re-summarizing conflicting text responses

Innovation

• Decomposes draft synthesis into atomic claims
• Scores sources with multi-task peer-prediction mechanism
• Filters unreliable sources before final re-summarizing