Co-occurrence Graph Convolutional Networks with Approximate Entailment for knowledge graph embedding

The goal of knowledge graph completion (KGC) is to address the issue of missing triples and enhance the overall completeness of the knowledge graph. However, existing methods face three key challenges: (1) Weak semantic correlation between entities and relations in the knowledge graph. (2) Insufficient extraction of local features in the model. (3) Limited ability to represent complex semantic relations. This paper proposes a G raph C onvolutional N etwork framework that leverages C o-occurrence features, local structural features, and A pproximate E ntailment (CoAE-GCN). The CoAE-GCN model is designed to overcome these challenges. The CoAE-GCN model addresses these issues by (1) enumerating the co-occurrence of entities and relations and using resulting weighted information as input for the model. (2) We employ Graph Neural Networks (GNNs) to learn structural features while using attention mechanisms to capture local structural features from incoming and outgoing neighbors. (3) We are applying approximate entailment to enhance the representational capacity of relations. Experimental results on benchmark datasets demonstrate that the CoAE-GCN model is outperformance and effective.