Cross-Knowledge Graph Entity Alignment via Neural Tensor Network

Abstract

AbstractWith the expansion of the current knowledge graph scale and the increase of the number of entities, a large number of knowledge graphs express the same entity in different ways, so the importance of knowledge graph fusion is increasingly manifested. Traditional entity alignment algorithms have limited application scope and low efficiency. This paper proposes an entity alignment method based on neural tensor network (NtnEA), which can obtain the inherent semantic information of text without being restricted by linguistic features and structural information, and without relying on string information. In the three cross-lingual language data sets DBPFR−EN, DBPZH−EN and DBPJP−EN of the DBP15K data set, Mean Reciprocal Rank and Hits@k are used as the alignment effect evaluation indicators for entity alignment tasks. Compared with the existing entity alignment methods of MTransE, IPTransE, AlignE and AVR-GCN, the Hit@10 values of the NtnEA method are 85.67, 79.20, and 78.93, and the MRR is 0.558, 0.511, and 0.499, which are better than traditional methods and improved 10.7% on average.