TimelineReasoner uses reasoning models to boost timeline summarization accuracy

The proliferation of online news makes extracting structured timelines from unstructured content challenging. While Large Language Models (LLMs) have been used for Timeline Summarization (TLS), they typically act as passive generators, unable to iteratively reason over events. To address this, researchers propose TimelineReasoner, a novel framework that leverages Large Reasoning Models (LRMs) to shift TLS from static generation to an active, reasoning-driven process. The framework operates in two stages: Global Cognition, which maintains a macroscopic view and continuously updates a global event memory, and Detail Exploration, which identifies informational gaps and refines the timeline through targeted document retrieval. Key components include an Event Scraper for retrieving temporal event descriptions, a Timeline Updater for refining the timeline, and a Supervisor for detecting gaps and guiding retrieval.

Experimental results on open-domain TLS datasets demonstrate that TimelineReasoner significantly outperforms existing LLM-based TLS methods in terms of timeline accuracy, coverage, and coherence. On closed-domain datasets, it performs on par with or exceeds state-of-the-art approaches. This work not only pushes the boundaries of TLS but also highlights the broader potential of LRM-based reasoning frameworks for timeline summarization. For professionals dealing with news aggregation or historical analysis, it promises automated generation of accurate, coherent timelines, saving significant manual effort.