Knowledge graph completion method by hierarchically aggregating multirelational neighborhood information

Knowledge graph is a heterogeneous graph containing multi-relational information. Knowledge graph embedding has become one of the most effective methods in knowledge graph completion task. Using Graph Attention Network as a knowledge graph embedding model can effectively improve the quality of knowledge graph completion. At present, the knowledge graph embedding model based on Graph Attention Network uses the single-layer attention mechanism to calculate the importance of all local neighborhood information of a entity in the knowledge graph. However, we observe that this method of placing all neighborhood information in the same layer to calculate attention weights ignores the heterogeneity of the knowledge graph. Therefore, we proposes a hierarchical aggregation method for multi-relational neighborhood information. The local neighborhood information of a entity is divided into a relationship layer and an entity layer, and the multi-layer attention mechanism is used to calculate its contribution to the entity. This method can aggregate the local neighborhood information of entity in a finer granularity, so that entity can aggregate more effective information and ignore more invalid information, thereby improving the performance of the model. In addition, our proposed model performs superior compared to several state-of-art methods.