基于OAN的知识图谱查询研究  

A Knowledge Graph Query Algorithm Based on OAN

作　　者：张玲玉 尹鸿峰[2] 

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044 [2]北京交通大学海滨学院计算机科学系,河北黄骅061199

出　　处：《软件》2018年第1期54-59,共6页Software

基　　金：河北省高等教育科技研究重点项目资助[ZD2017304]

摘　　要：知识图谱查询是目前知识图谱研究中最广泛的应用,能够有效提高搜索引擎查询效率。然而,现有的知识图谱的查询研究多是基于节点标签的子图匹配。由于节点标签不能体现节点间的语义信息,导致查询结果的语义相关性不高。针对此问题,本文提出了一种基于本体和邻居信息的查询算法OAN(Ontology and Neighborhood)。首先,结合本体相似度和邻居相似度来确定查询节点的候选集,以此提高候选节点的语义相似度;其次,通过边检测算法移除那些不满足条件的查询节点候选集,以此减少查询规模;然后,在目标图上查找满足边标签同构的查询子图,并计算节点的标签相似度和结构相似度总和,给每个结果集打分后排序,获得最终排序后的结果集;最后,通过在真实数据集上与已有查询算法进行对比实验,实验结果表明:本文所提出的方法无论是在精确度上,还是在查询效率方面都有所提高。Knowledge graph query is the most widely used in the current knowledge graph research and can effectively improve the query efficiency of search engine. Previous work on query technologies for knowledge graph has focused on the subgraph matching based on node label. However, node labels are difficult to present the semantic relationships between nodes, thus resulting in low semantic relevance for query results. In this paper, to improve the semantic relevance, we propose a graph querying algorithm called OAN （Ontology and Neighborhood） which combines both the ontology information and neighborhood information together during the search process. We identify the candidate set for query graph by measuring the node similarity and structural similarity. We further propose a pruning technique that removes the mismatched candidates to prune the search space. Finally, we find matching results from target graph by edge-label isomorphic querying and rank the top-k graph matches by their similarity. Experiments on real-life datasets show that OAN can significantly improve the precision of results, and outperforms the state-of-the-art graph querying methods on both flexibility and scalability.

关 键 词：知识图谱 子图匹配 本体 邻居节点 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]