KGU-SP:一种挖掘标准模式的知识图谱更新方法

KGU-SP:Knowledge Graph Updating Method Using Standard Pattern

目前知识图谱更新面临着更新知识数量大、频率高且包含大量错误信息的问题,现存的方法为保证更新效率对知识简单分类,判断方法单一且不考虑知识的语义关联,导致知识更新精确度低.本文提出一种基于标准模式挖掘的知识图谱更新方法,该方法与自动抽取并直接插入的方法不同,利用从现有知识图谱中挖掘出的标准模式进行模拟匹配来筛选正确的知识,并对标准模式挖掘算法进行了改进,提升了筛选的准确性.当大量知识同时进行更新时对挖掘出的标准模式和待更新知识同时向不同的领域空间向量化,计算两者间的向量距离,通过控制阈值筛选正确的知识.在真实的数据集上实验结果证明了本方法的准确性和有效性.

At present,the updating of knowledge graph faces the problems of large quantity,high frequency and large amount of error information.In order to ensure the updating efficiency,the existing methods simply classify the knowledge.The judgment method is simple which consider the semantic association of knowledge,thus leads to the low accuracy of knowledge updating.This paper proposes a new method of knowledge graph updating based on standard pattern mining.Different from the method of automatic extraction and direct insertion,it uses the standard patterns mined from the existing knowledge graph.When a large amount of knowledge is updated at the same time,the standard pattern and the knowledge to be updated are vectorized to different domains at the same time,and the vector distance between them is calculated,by controlling the threshold to filter out the correct knowledge.The experimental results on real data sets prove the accuracy and effectiveness of this method.