关键词: 知识图谱, 补全, 用户生成数据, 互信息, 关联影响

Abstract:

The completion of missing relationships between entities in knowledge graph (KG) is the topic with great attention in the field of KG research. With the rapid development of Web2.0, the association between entities reflected by the user-generated data (UGD) is complementary to the knowledge described in KG. In the knowledge reasoning method based on KG path, there are sparse or wrong entity relations and poor connectivity, which leads to the inaccurate relationship extracted from entities. For this problem, this paper proposes a method for complementing KG by using correlation between entities in UGD. Firstly, based on the UGD, this paper uses mutual information to calculate the relationship between entity nodes and build the entity association graph (EAG), and then proposes a superposition method to quantify the potential correlation between non-adjacent entities in the EAG, so the association impact values are obtained. Finally, the multiple correlation effects between non-adjacent entity nodes are superposed to determine whether there is a strong correlation between the entities. By adding the edges between non-adjacent entity nodes with associations, KG completion can be fulfilled. The experimental results based on real data sets show the efficiency and effectiveness of the proposed KG completion.

Key words: knowledge graph, completion, user-generated data, mutual information, association impact