A Lightweight GNN-Based Graph Embedding Method for Knowledge Graphs
Weide Huang, Linlan Liu, Jian Shu · 2024
Knowledge graph is a graphical knowledge base for organizing and representing knowledge. In knowledge graph, each piece of knowledge is represented as a triplet form of (head entity, relation, tail entity). The main idea of graph embedding based on graph neural networks is to project the entities and relations in the knowledge graph into a low-dimensional space for various downstream tasks. However, existing GNN-based knowledge graph embedding methods suffer from the problem of excessive number of parameters. To solve the problem, we propose a lightweight GNN-based knowledge graph embedding method. Lightweight entity feature vectors are obtained by splicing entities and relations. We validate the effectiveness of our method on a link prediction task. Comparisons are made with three different baseline methods on two real-world datasets Kinship and UMLS. The results show that our embedding method reduces the number of parameters in the model without loss of accuracy.