融合知识图谱和语义匹配的医疗问答系统OACSTPCD

Question answering system is an important task in the field of natural language processing,which is often used in medical service.The traditional question answering system can return the corresponding tail entity as the answer by means of the entity and relationship matching of the knowledge graph.However,if the entity or relationship is not recognized or there is no corresponding entity relationship in the knowledge graph,the question answering can not be continued.In order to solve this problem,a hybrid framework of Chinese medical question answering is proposed,which combines knowledge graph and semantic matching model.When the questions raised cannot be matched by the entity relationship in knowledge graph,the model can continue to find the most similar questions from the question answering on datasets and return corresponding results as answers.In terms of semantic matching models,combining Chinese medical similarity problems,fine-tuning training is conducted on the Sentence BERT model,and distance measurement functions in hyperbolic space are introduced to measure the sentence similarity.The results show that in terms of overall performance,the proposed model can improve accuracy by 7.16%compared to large language models like BERT.In terms of measurement ability,in combintion with the general Euclidean space metrics such as cosine metrics,hyperbolic metrics can achieve a maximum accuracy improvement of 2.28%and an F1 value improvement of 1.58%.