Reproduce “AI Historian: Helping historians organize and verify person-centred temporal clues from dispersed historical narratives” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yifeng Lu, Zijie Yang, Jie Li, Qingkai Min, Yue Zhang
Researchers from Westlake University and Peking University developed AI Historian (AIH), a multi-agent AI system designed to organize and verify person-centered temporal information from fragmented historical texts. The system achieved an 86.2% MicroIoU in temporal reconstruction, outperforming human-only annotation at 81.3% and reducing processing time from 92 minutes to 14 minutes for specific *Shiji* cases.
Key insights
- Decomposing complex historical temporal reconstruction into a multi-stage, inspectable AI agent workflow significantly improves both accuracy and traceability compared to direct LLM application or manual efforts.
- Structured prompting, mimicking the decomposed workflow, particularly enhances cross-text event verification and temporal alignment, mitigating inherent limitations of large language models.
- Establishing a common, item-level traceable representation for temporal evidence allows for systematic source criticism and cross-textual comparison across disparate historical narratives.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.