A Novel Embedding Model for Knowledge Graph Completion Based on Quaternion

Abstract

In recent years, knowledge graph completion methods have been extensively studied, in which QuatE learned embeddings of entities and relations in quaternion space and achieved state-of-the-art results. However, QuatE has two main problems: 1) simple modeling operation leads to weak interaction between entities and relations and inflexible representation. 2) complex relations are not to be considered. In this paper, we propose a novel model, en-QuatE, with a dynamic mapping strategy to explicitly capture a variety of relational patterns, enhancing the feature interaction capability between elements of the triplet. The mapping strategy dynamically, associated with the relation, used to learn adaptive the entity embedding vectors in the quaternion space via Hamilton product. Experiment results show en-QuatE achieves significant performance on WNISRR. In particular, the MR (Mean Rank) evaluation has relatively increased by 15% on WNISRR.