| Currently,knowledge graphs are widely used in different domain scenarios,and the size and data representation of knowledge graphs may vary due to different specification standards and design requirements.And the information provided by a single knowledge graph may not be sufficient to meet the task requirements,so entity alignment is usually considered to fuse multiple knowledge graphs into a large unified knowledge graph.The existing entity alignment research methods mainly use the structural information of knowledge graphs to determine whether the entity pair is aligned or not by measuring the distance between two entities.However,there are usually heterogeneous redundancies in different knowledge graphs,and most entities in the real world are long-tailed entities,so relying only on structural information will lead to low entity vector expressiveness and insufficient information propagation,which directly affects the performance of entity alignment.To address the above problems,this paper proposes a multi-feature fusion-based entity alignment model that combines different entity feature information to improve the accuracy of entity alignment,and the specific research work and contents are as follows.(1)Firstly,this paper uses the Rotat E model to embed the structural information of the knowledge graph into the complex space,combining the vector representation of entities taking different roles to improve the representation of entity embedding.Secondly,a pre-trained word embedding BERT model is used to obtain the initial vector representations of entity names and attribute features,respectively,and a multilayer graph neural network is used to capture the neighboring entity information to obtain more expressive name embeddings.In addition,attribute interaction features between modeled entities based on attribute value embeddings are used to make full use of the semantic information of the knowledge graph.Finally,the entity similarity matrices of structure,name,and attributes are dynamically fused to generate a similarity matrix to measure the plausibility of the alignment of the two entities.Comparative experiments are conducted on three cross-lingual datasets of DBP15 K,and the experiments show that the model designed in this paper outperforms most other models and can effectively improve entity alignment accuracy.(2)This paper further introduces knowledge graph complementation to solve the problem of long-tailed entities.By combining entity alignment and knowledge map complementation to expand the structural information of the knowledge map,iterative training enhances the performance of knowledge map complementation and entity alignment and further improves the coverage rate in sparse knowledge maps.In this section,experiments are conducted on the SRPRS dataset,and it is demonstrated that the combination of knowledge graph complementation and entity alignment can effectively increase the number of triples and reduce the number of long-tailed entities. |
