MulEA: Multi-type Entity Alignment of Heterogeneous Medical Knowledge Graphs.

The large-scale application of medical knowledge graphs has greatly raised the intelligence level of modern medicine. Considering that entity references between multiple medical knowledge graphs can lead to redundancy, knowledge graph alignment tasks are required to identify entity pairs or subgraphs of heterogeneous knowledge graphs pointing to the same elements in the real world. Existing medical knowledge graph alignment methods do not consider the multi-type features of medical entities. To tackle above challenge, we propose a Multi-type Entity Alignment model of medical knowledge graphs based on attention mechanism named MulEA. Firstly, MulEA integrates the multi-type information features of medical entities to align various types of entities jointly through constructing an entity multi-type information embedding matrix for medical knowledge graphs. Secondly, MulEA designs a relationship collective aggregation module to fully utilize the features of different relationships to improve alignment accuracy. Finally, we evaluate the performance of MulEA on a real medical knowledge graph alignment dataset MED-OWN-15K, and it achieves the state-of-the-art performance on several metrics, showing the effectiveness of our method.