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

The paper introduces NTS-CoT, a novel framework designed to address hallucinations in LLM-based timeline summarization (TLS) by employing Chain-of-Thought (CoT) reasoning. It comprises three modules: Element-CoT for capturing essential news elements, Date Selection for optimizing timestamp selection, and Causal-CoT for inferring causal relationships. Experimental results show that NTS-CoT significantly outperforms existing benchmarks in reducing hallucinations and enhancing summarization accuracy, which is critical for practitioners aiming to improve the reliability of LLM-generated news content.