天津科技大学学报2023，Vol.38Issue(6)：54-61,8.DOI:10.13364/j.issn.1672-6510.20230016
知识嵌入的医疗对话生成
Knowledge-Embedded Medical Dialogue Generation
摘要
Abstract
Aiming at the problem of lack of medical common senses consistency caused by the failure of medical knowl-edge modeling in previous medical conversation generation methods,many researchers have tried to introduce the medical knowledge graph,but it was easy to occupy too much input data space when integrating the medical knowledge graph,which has limited the dialogue context information that could be retained by the model input.In this article,a medical conversation generation model based on knowledge embedding(MCG-KE)is proposed.This model makes entity prediction based on historical dialogue to obtain context knowledge embedded entity,and introduces serial graph coding and graph attention mechanism to obtain the subgraph coding of medical knowledge graph related to current dialogue.Context knowledge em-bedding entity,medical knowledge spectrum subgraph coding and historical dialogue coding are used as input of dialogue generation model for knowledge embedding medical conversation generation.The experimental results showed that,under the condition of efficient calculation,the performance of the medical conversation generated by the model was improved in the relevant indexes such as automatic evaluation and manual evaluation.关键词
医疗对话生成/上下文知识嵌入/知识图谱子图编码/图注意力机制/生成模型Key words
medical conversation generation/contextual knowledge embedding/knowledge graph spectrum subgraph coding/graph attention mechanism/generation model分类
信息技术与安全科学引用本文复制引用
王嫄,曾磊磊,武振华,熊宁..知识嵌入的医疗对话生成[J].天津科技大学学报,2023,38(6):54-61,8.基金项目
国家自然科学基金项目(61976156) （61976156）
大学生创新创业训练计划项目(202210057063) （202210057063）
