融合机器阅读理解的知识图谱问答系统设计与实现

Design and Implementation of Knowledge Graph Answering System Combining Machine Reading Comprehension

智能问答系统是信息检索的一种新型检索模式。针对现有知识图谱问答系统存在的开放域知识覆盖不全而影响问答质量的问题,提出了一种融合机器阅读理解的知识图谱问答系统构建方法。首先采用Elasticsearch数据库来存储预处理之后的三元组数据,其次利用信息检索技术在知识图谱中对问题相关知识进行检索,并将检索结果拼接成一段自然语言描述,采用XLNet模型进行答案抽取匹配,最后在2019年度第二届中国“AI+”创新创业大赛-自然语言处理技术创新大赛的数据集上进行了实验,其F1值能够达到0.23的指标。实验证明,融合机器阅读理解模型的知识图谱系统在复杂的开放域中文问答任务具有更好的表现能力。

Intelligent question answering system is a new retrieval model of information retrieval.Since incomplete open-domain knowledge coverage affects system quality,we propose a method of constructing knowledge graph question answering system based on machine reading comprehension.Firstly,we use the Elasticsearch database to store the preprocessed triples.Secondly,we retrieve the knowledge related to the question in the knowledge graph with information retrieval technology,and the retrieval results are spliced into a natural language description.Then the answers are extracted and matched with the XLNet model.Finally,the experiment is carried out on the dataset of the Second Chinese"AI+"Innovation and Entrepreneurship Competition Natural Language Processing Technology Innovation Competition in 2019,with its F1-score reaching 0.23.Experiments show that the knowledge graph questioning answer system based on machine reading comprehension has better performance in complex open domain Chinese question answering tasks.