摘要
提出了一种基于Seq2seq半指针半标注法的对输电规程文本实体关系进行联合抽取方法,构建了电力知识图谱。该方法首先使用指针网络解码提高实体抽取的准确性解决了关系抽取中实体边界模糊问题,共享编码层进行关系抽取解决了串联抽取方法误差积累的问题;然后采用查询匹配系数法,优化了三元组数据集的筛选与排序,提高了电力知识图谱的查全率和查准率。实验结果表明,在人工标注的电力领域关系数据库中测试集F1值达到0.837 6,在知识图谱查询测试实验中平均查准率和查全率达到了75%以上,验证了该方法在输电规程文本抽取的有效性,最后将抽取到的电力知识图谱以三元组形式存入图数据库Neo4j,实现数据查询可视化,为电网规程文本的准确查询提供了依据。
This paper proposes a method for joint extraction of entity relations in the text of power transmission regulations based on the Seq2 seq semi-pointer and semi-annotation method,thereby we construct a power knowledge graph.The method first uses pointer network decoding to improve the accuracy of entity extraction,and solves the problem of entity boundary ambiguity in relation extraction.The shared coding layer performs relation extraction to solve the problem of error accumulation in the concatenated extraction method.Then,the query matching coefficient method is used to optimize the ternary The screening and sorting of group data sets improves the recall and precision of the power knowledge graph.The experimental results show that the F1 value of the test set reaches 0.837 6 in the manually labeled relational database in the power field,and the average precision and recall rate in the knowledge graph query test experiment reaches more than 75%.Finally,the extracted power knowledge graph is stored in the graph database Neo4 j in the form of triples to realize the visualization of data query and provide a basis for accurate query of power grid regulations text.
作者
何俊
刘鹏
聂勇
吴慎珂
刘鹏政
钟可佳
HE Jun;LIU Peng;NIE Yong;WU Shenke;LIU Pengzheng;ZHONG Kejia(College of Information&Engineering,Nanchang University,Nanchang 330031,China)
出处
《实验室研究与探索》
CAS
北大核心
2022年第7期1-5,17,共6页
Research and Exploration In Laboratory
基金
国家自然科学基金项目(62066025)
教育部产学合作协同育人项目(202102607005)。
关键词
电力知识图谱
实体关系联合抽取
知识查询
electricity knowledge graph
entity relationship joint extraction
knowledge query