Knowledge Graph Embedding Based on Triple Multilayer Perceptron

The knowledge graph is a structured semantic knowledge base, that can organize the information in the Internet at the minimum cost and can be applied to many downstream fields at the same time. However, knowledge graphs in the real world are mostly incomplete, and the idea of knowledge graph embedding is proposed to solve this problem. In the early days, the translation series model was used for knowledge graph embedding, but it was only suitable for modelling some simple relations. The neural network-based model has more advantages in capturing complex features, so it has more advantages in link prediction than traditional translation models. This paper proposes a model called TMLPE (Triple Multiplayer Perceptron Embedding), the main body of which is composed of three MLP modules, one of which is used to capture the interaction features between head entities and relations, and one MLP model is used to capture the characteristics of the head entities and relations themselves. An MLP is used to project the tail entities into a vector space similar to that of the head entities and relations. Experiments show that TMLPE exhibits good link prediction performance on two benchmark datasets.