基于医疗知识图谱的智能问答系统研究

Intelligent Question Answering System Based on Medical Knowledge Graph

【目的/意义】研究融合知识图谱中医疗知识解析用户问句中的命名实体和关系,提升智能医疗问答系统对用户问句语义解析能力,为用户提供更有效的自助医疗问答服务。【方法/过程】首先采集医疗知识构建医疗知识图谱,再利用图谱中的知识强化基于多重注意力机制的命名实体识别模型以解析医疗问句中的实体,然后采用基于BERT-BiLSTM的关系抽取模型进行关系抽取,最后利用解析结果生成查询语句从知识图谱中获取问题的答案。【结果/结论】通过对比实验将本文设计的语义解析模型与智能问答中常用的其他模型在医疗问句数据集上进行比较,发现准确率、召回率和F1值均有所提升,验证了本文智能问答系统对用户问句理解更准确。【创新/局限】本文构建了医疗知识图谱,并将图谱知识与基于多重注意力机制的语义解析模型相结合构建了医疗智能问答系统,在问答任务中具有较好的表现,在复杂问句的解析上还有待进一步研究。

【Purpose/significance】Research on fusing the medical knowledge in knowledge graph to analyze the named entities and relationships in user questions and to improve the semantic parsing ability of the intelligent medical question answering system,so as to provide users with more effective self-help medical question answering services.【Method/process】Medical knowledge is firstly collected to build medical knowledge graph,then the knowledge in the graph is used to strengthen the named-entity recognition model based on multi-attention mechanism to analyze the entities in medical questions,and the relationship extraction model based on the BERT-BiLSTM to extract relations,finally,the analysis results are used to generate query statements and obtain answers from the knowledge graph.【Result/conclusion】Through comparative experiments,the semantic parsing model designed in this paper is compared with other models commonly used in intelligent question answering on the medical question datasets,and it is found that the accuracy rate,recall rate and F1 score are improved,which verifies that the intelligent question answering system in this paper can understand user questions more accurately.【Innovation/limitation】This paper constructs a medical knowledge graph and a question answering system combining the graph knowledge and the semantic parsing model based on multi-attention mechanism,which has better performance in question answering tasks,and the parsing of complex questions needs further research.