Attention-Based Relation Prediction of Knowledge Graph by Incorporating Graph and Context Features

Abstract

As the vital resource for various applications like question answering and recommendation system, knowledge graph (KG) often suffers from incompleteness. The task of relation prediction in KG aims to infer the relations between entities, which depend on the structure of the query-specific subgraph but also the neighborhood as context. In this paper, we propose an attention-based joint model for relation prediction by incorporating the graph and context features of entities and relations. First, we extract the subgraph and entity context from KG, and adopt the attention mechanism to capture the most relevant graph and context features, which we leverage to generate the enhanced representations. Then, we give the joint loss function to guarantee the unified representation including the relevant graph and context features simultaneously. Finally, we fulfill model training for relation prediction. Experimental results on real-world datasets demonstrate that our proposed approach outperforms the state-of-the-art competitors.