A Semantic Filter Based on Relations for Knowledge Graph Completion

• DOI: 10.18653/v1/2021.emnlp-main.625
• Publication Date: 2021-12-17T03:56:42Z
• Authors: Zongwei Liang; Junan Yang; Hui Liu; Keju Huang
• Keywords: 01 natural sciences; 0105 earth and related environmental sciences

Knowledge graph embedding, representing entities and relations in the knowledge graphs with high-dimensional vectors, has made significant progress link prediction. More researchers have explored representational capabilities of models recent years. That is, they investigate better to fit symmetry/antisymmetry combination relationships. The current embedding are more inclined utilize identical vector for same entity various triples measure matching performance. observation that measuring rationality specific means comparing degree attributes associated is well-known. Inspired by this fact, paper designs Semantic Filter Based on Relations(SFBR) extract required entities. Then compared under these extracted through traditional models. semantic filter module can be added most geometric tensor decomposition minimal additional memory. experiments benchmark datasets show based suppress impact other attribute dimensions improve prediction SFBR achieved state-of-the-art.