NTS-CoT: Mitigating Hallucinations in LLM-based News Timeline Summarization with Chain-of-Thought Reasoning

2026-06-11

AI Summary

This work addresses the susceptibility of large language models to two types of hallucinations in timeline summarization (TLS)—content distortion and omission of date-event pairs—and proposes NTS-CoT, the first framework to systematically identify and mitigate such hallucinations in TLS. NTS-CoT integrates Element-CoT and Causal-CoT mechanisms, leveraging chain-of-thought reasoning to jointly optimize the extraction of key news elements, modeling of temporal dynamics and event salience, and causal inference. Experimental results on three TLS benchmarks demonstrate that NTS-CoT significantly outperforms existing methods, with both human and automatic evaluations confirming its effectiveness in reducing hallucinations and enhancing summary fidelity and completeness.

Abstract

The rapid updates of online news make tracking event developments challenging, highlighting the need for timeline summarization (TLS). Hallucinations, where LLM-generated content deviates from source news, still remain a critical issue in LLM-based TLS and are not well studied in existing works. To bridge this gap, we identify two primary types of hallucinations: unfaithful content during news summarization and information omission in date-event summarization. Then, we propose NTS-CoT, a novel framework that leverages Chain-of-Thought (CoT) reasoning to mitigate hallucinations in TLS. The framework consists of three key modules: i) Element-CoT to capture essential news elements for faithful summarization, ii) Date Selection to combine temporal saliency and event prominence for timestamp selection, and iii) Causal-CoT to infer causal relationships and reduce omissions in date-event summarization. Extensive experiments, including quantitative analysis on three TLS benchmarks and human evaluation, demonstrate that NTS-CoT outperforms state-of-the-art baselines, effectively mitigating hallucinations and improving LLM-based TLS performance. Our source code is available at https://anonymous.4open.science/r/NTS-CoT .

Problem

• hallucinations
• timeline summarization
• LLM-based summarization
• news summarization
• faithfulness

Innovation

• Chain-of-Thought Reasoning
• Hallucination Mitigation
• Timeline Summarization
• News Summarization
• Causal Inference