Design and Implementation of an Intelligent Ancient Chinese Poetry Search System Based on LLMs
Wofeng Chen, River Dai, Kai Chen · 2024
This paper explores the design of an intelligent ancient Chinese poetry search system using large language models (LLMs) to address limitations in existing methods regarding search intent, authority, and completeness. We built a dataset of 220,000 poems with structured information, designed a two-stage system leveraging LLMs for semantic extraction and query transformation, and proposed prompt engineering techniques combining chain-of-thought reasoning and external knowledge. A prototype achieved 75% accuracy, demonstrating effective intent understanding and semantic extraction without custom training. This study lays a foundation for intelligent poetry search systems, with all code and datasets available on GitHub.