面向知识扩充的实体关系挖掘

摘要

为了从自由文本中挖掘大量高质量的事实抽象出三元组，并将其与现有的知识库进行知识融合，提出了知识图谱构建中的知识扩充框架。首先对知识图谱的构建过程进行了认知，指出传统的关系挖掘仅仅是利用句法依存树抽取路径上的节点作为关系，对于复杂句子表现较差。在此基础上建立了一种基于关系类型的结合多种特征的知识扩充框架，该方法自动获取高质量知识。效果在开放网页句子中达到F1值88%。%To extract the high quality facts as triplets and merge them into existing knowledge base, an extended frame of knowl-edge growth in the construction of knowledge base is presented. Firstly, the process of knowledge base constructing is recog-nized. It is pointed out that the limitation in recognizing relationship between entities is to only rely on parsing dependency tree, and perform worse on complex sentences which contain large amount of entities. Then, the knowledge growth framework fusing different types of features is proposed for extracting high quality knowledge automatically. Finally, in web data sentences, the F1 measure of 88%of the presented method is demonstrated.