基于道路法规知识图谱的多轮自动问答研究  

Research on Multi-round Automatic Question Answering Based on Knowledge Graph of Road Regulations

作　　者：陈金菊[1] 王义真 欧石燕[1] Chen Jinju;Wang Yizhen;Ou Shiyan(School of Information Management,Nanjing University,Nanjing 210023,China)

机构地区：[1]南京大学信息管理学院,江苏南京210023

出　　处：《现代情报》2020年第8期98-110,120,共14页Journal of Modern Information

基　　金：国家社会科学基金重点项目“基于关联数据的学术文献内容语义发布及其应用研究”(项目编号:17ATQ001)。

摘　　要：[目的/意义]传统的基于知识图谱的自动问答研究主要是针对用户提问直接检索答案,但由于系统对用户问题的理解存在歧义,导致得到的答案不够精确。采用基于知识图谱的多轮自动问答可以有效地改善这一问题。[方法/过程]本文首先构建了以事件为中心的道路法规本体模型,依据该模型从道路法规中抽取实例图谱,并设计出基于道路法规知识图谱的问答框架。然后,对该框架所使用到的模型进行测评。最后,进行系统的总体测评。[结果/结论]从模块测评结果来看,本文所提出BCNN_BiLSM模型在事件识别和意图识别的F1值分别是0.798和0.930,BBiLSTM_CRF模型在本体属性识别F1值为0.807,总体性能优于其他模型。系统的总体测评结果表明,完整句的准确率为0.74,缺省句的任务完成率为0.83。本文提出的基于道路法规知识图谱的多轮自动问答可为相关领域自动问答研究提供参考。[Purpose/Significance]The traditional researches on knowledge-based automatic question answering aim to retrieve answers for user questions directly.However,due to the ambiguity of the system's understanding of the questions proposed by users,the answers obtained are not accurate enough.The use of multiple-round of automatic question answering based on knowledge graph can effectively improve this situation.[Method/Process]This paper firstly constructed an event-centered road regulation ontology model,based on which the sample graphs are extracted from the road regulations.And a question answering framework based on the knowledge graph of road regulations is designed.Then the model used in the framework was evaluated.Finally,an overall evaluation of the system was carried out.[Result/Conclusion]From the results of the module evaluation,the F1 values of the event identification and intent recognition of the BCNN_BiLSM model in this paper were 0.798 and 0.930 respectively,and the F1 value of the ontology attribute identification of the BBiLSTM_CRF model in this paper was 0.807,and the overall performance was better than other models.The overall evaluation results of the system showed that the accuracy of the complete sentence was 0.74,and the task completion rate of the default sentence was 0.83.The multi-round automatic question answering based on the knowledge graph of road regulations proposed in this paper can provide reference for the automatic question answering studies in related fields.

关 键 词：知识图谱 本体 多轮自动问答 自然语言处理 道路法规 

分 类 号：G250.7[文化科学—图书馆学]