基于重要性筛选与语义聚类的本体摘要技术  

Ontology Summarization Based on Importance Filtering and Semantic Clustering

作　　者：吴子仪 李邵梅[1] 张建朋[1] 姜梦函 WU Ziyi;LI Shaomei;ZHANG Jianpeng;JIANG Menghan(Information Engineering University, Zhengzhou 450001, China)

机构地区：[1]信息工程大学,河南郑州4500001

出　　处：《信息工程大学学报》2021年第6期716-721,共6页Journal of Information Engineering University

基　　金：国家自然科学基金青年科学基金资助项目(62002384);郑州市协同创新重大专项资助项目(162/32410218)。

摘　　要：本体是知识图谱的上层结构,为了帮助用户快速、灵活地理解和使用本体,本体摘要技术应运而生。针对现有的本体摘要技术中对语义信息利用不充分以及易向长文本倾斜的不足,提出基于重要性初筛和语义聚类的两级本体摘要方法。首先,基于预定义的重要性度量指标筛选候选本体概念集;其次,采用BERT将每个候选本体概念转化为对应的本体概念向量;最后,使用K-means++聚类对这些本体概念向量进行聚类,获得具有代表性的重要概念,完成本体摘要。实验结果表明,相对于已有方法,提出的方法可以挑选出更具代表性的本体概念。As the superstructure of knowledge graph,ontology is the normalized and formalized description of abstract concepts and relations in a certain domain.In order to understand ontology quickly and flexibly,ontology summarization arises.To solve the problem of the insufficient utilization of semantic information and the tilt toward long text,a two-level ontology summarization based on importance filtering and semantic clustering is proposed.First,the candidate ontology concept sets are filtered by predefined importance measures.Then,BERT is used to transform each candidate ontology concept into its corresponding ontology concept vector.Finally,K-means++is used to cluster these ontology concept vectors to obtain representative concepts.The experimental results show that compared with the existing methods,the proposed ontology summarization method can pick out more representative ontology concepts.

关 键 词：知识图谱 本体摘要 BERT K-means++ 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]