Multi-Relational Graph Convolutional Network Based on Relational Correlation for Link Prediction

Abstract

Knowledge graphs connect different entities through relationships, Multi-relational knowledge graphs are the common graph form. There are many unexplored potential relationships in multi-relational knowledge graphs. Link prediction is commonly used for knowledge graph completion, The link prediction task can infer possible relationships based on existing entities. Inspired by the advances of graph convolutional networks the link prediction task, we proposed a relational relevance-based GCN framework called RC-CompGCN. Firstly, update the embedding of all low-dimensional relations using the relational correlation module. Secondly, combined embedding entities and relationships using the graph structure module and various entities in knowledge graph embedding techniques are utilized relationship combination operations. We use the relational correlation module and graph convolutional network for link prediction tasks for the first time.