Construction and optimization of medical question answering system based on Knowledge Graph

Abstract

Medical question answering system is one of the significant measures to effectively solve common sense queries related to diseases. Aiming at the loose structure, low efficiency, and weak linking of domain knowledge in the current traditional medical model field, this paper constructs a Medical question answering system based on knowledge graph(KG). We achieve the goal of improving classification performance through three aspects, including: (i) creation of a medical KG by building a dictionary and generating entities and relationships, (ii) design of the Bert+TextCnn model combined with the k-nrm text matching algorithm to calculate the question template that is closest to the user's question in the system, and (iii) construct a cypher sentence based on the medical entity and the question template. The experimental results demonstrate that the medical KG question answering system based on intent recognition and k-nrm text matching has a higher accuracy rate than the traditional question answering system, and has a extraordinary question answering ability.