Research On The Method Of Entity Alignment With Attribute Enhancement Based On Graph Convolution

The entity alignment of knowledge graph is helpful to the creation and expansion of knowledge graph,and then promote the application and development of knowledge graph.The method that relies on knowledge graph experts to manually annotate entity alignment is time-consuming and has some errors.In the current stage of rapid development of deep learning,it relies on excellent network structure and good learning ability and has good performance on knowledge graph.In this paper,the task of entity alignment is studied based on deep learning method.The main work and contributions of this paper are as follows:(1)The entity alignment method GCN-align of graph convolutional neural network was supplemented and analyzed in detail.On dataset DPK15 K,the entity alignment effect of attribute structure was analyzed separately.Attribute structure and relation structure are interchanged and spliced to analyze the parameter effect change of splicing weight.At the same time,attribute structure and relation structure are analyzed together.Finally,the experimental results on another dataset DWY100 K are given.Experimental results show that the effect of graph convolution on the alignment of relational structure is more significant than that of attribute structure,and the auxiliary effect of attribute structure on the alignment of relational structure can effectively improve the alignment effect.At the same time,comparing data sets of different sizes,it is found that the larger the size of data sets,the worse the effect of entity alignment.(2)According to the characteristics of the knowledge map and entities,and the relationship between research heterogeneous knowledge map awareness entity alignment method,for further improved,and put forward the combination of attribute information of heterogeneous knowledge map relation perception entity alignment methods,better realize the different language knowledge map background,the same entity to align,introducing auxiliary entity attribute information alignment tasks.Experimental results show that the proposed method is simple in thought and implementation,and can improve about 2% on@1 and 4% on @10 and @50,respectively,compared with the relational aware entity alignment method of heterogeneous knowledge graph.The accuracy of the proposed method is further improved.(3)Based on the relationship awareness neighborhood matching model(RNM)based on entity alignment,an entity alignment relationship awareness neighborhood matching model combining attribute and dual attention mechanism was proposed.The dual attention of RDGCN was introduced to optimize the relational structure learning ability of GCN.Add attribute information,combine the relation structure and attribute information as relational aware neighborhood matching embedding.The accuracy of alignment on ZH-EN,JA-EN and FR-EN of DBP15 K is 86.91%,87.67% and 94.05%,respectively,which is further improved compared with the benchmark model.Experimental results show that this method can identify the aligned entity pairs more effectively.