FinQA: A Training-Free Dynamic Knowledge Graph Question Answering System in Finance with LLM-Based Revision

Wenbiao Tao, Hanlun Zhu, Keren Tan, Jiani Wang, Yuanyuan Liang, Huihui Jiang, Pengcheng Yuan, Yunshi Lan

Abstract: Knowledge graph question answering (KGQA) in the finance domain aims to answer questions based on a dynamic knowledge graph (KG), which suffers from frequent updates. Moreover, the lack of high-quality annotated data renders data-driven and training-dependent approaches ineffective. To bridge the gap, we develop FinQA, which is a training-free dynamic knowledge graph question answering system in finance with large language model based (LLM-based) revision. Specifically, FinQA gives considerations to the following aspects: (1) constructing a dynamic finance knowledge graph partitioned based on data update frequencies; (2) proposing a training-free question-answering (QA) system to parse natural language to graph query language (NL2GQL) and achieving high-efficient coordination with the dynamic KG; (3) integrating the QA system with an open-source LLM to further boost the accuracy.