Research on Alignment Method of Civil Aviation Equipment Domain and General Knowledge Graph Entity Based on Graph Neural Network Embedding
Dongdong Guo, Tianyu Wu, Anna Fu · 2023
Because of the lack of description information, significant attribute differences and substantial relationship heterogeneity existing in the alignment of domain knowledge graphs and general knowledge graphs, the attribute information is embedded through the graph attention mechanism, the entity representation vector obtained by attribute embedding is used as the initial vector of relation embedding, and the graph attention network is used for relation embedding, an entity representation vector containing attribute and relationship information is obtained. The entity representation vector is aligned by similarity calculation. Experimental results on the alignment dataset CAKG-ONTK demonstrate that the proposed model outperforms other typical graph neural network-based entity alignment models. To validate the effectiveness of the relationship embedding method used in this paper, experiments are conducted using GCN, R-GCN, and GAT, which are graph neural network models. The results show that the relationship embedding method proposed in this paper achieves the best performance. Furthermore, to assess the robustness of the model, experiments are conducted using different proportions of entity seed pairs, as well as different hyperparameters. The results demonstrate that the model exhibits strong robustness. Compared with the mainstream alignment model based on graph neural networks, this model has a specific improvement in the Hist@1 and Hist@10 on the domain knowledge graph and general knowledge graph alignment dataset.