Research Of Semi-supervised Knowledge Graph Entity Alignment Algorithm Based On Transfer Learning

With the rapid development of the Internet,the volume of network data is increasing.The knowledge graph is used to better describle and organize the massive Internet data.The knowledge graph is a semantic network that describes entities,concepts and their associations that exist objectively in the real world.The knowledge graph is connected with various entities through various relations.However,most knowledge graphs are still very sparse and incomplete,which seriously affects the development of downstream applications driven by knowledge graphs.In order to solve this problem,entity alignment is used to find entities that point to the same thing in the real world between the current and other external knowledge graphs.Then,the relations and attributes information in other aligned graph can be further used to expand and complete the current knowledge graph.In fact,existing entity alignment algorithms mostly focus on semisupervised methods to solve the problems of limited alignment labels and cost of obtaining labeled data between different knowledge graphs,but the alignment effect is not good.Transfer learning can extract useful information from one or more source domain tasks and apply it to the new target domian tasks.Based on the transfer learning method,the accuracy of semi-supervised entity alignment model will be effectively improved by transferring common alignment knowledge of entity alignment tasks between different knowledge graphs.Therefore,based on the method of transfer learning,this thesis explores the semi-supervised entity alignment algorithm of two task scenarios.One is the source domain has sufficient alignment labels and the target domain has few alignment labels.And the other task scenario is that both the source domain and the target domain have a small number of alignment labels.Aiming at the entity alignment task in the scenario where the source domain has sufficient alignment labels and the target domain has few alignment labels,this thesis proposes a semi-supervised entity alignment algorithm based on domain adaptation learning.Firstly,in order to improve the entity representation quality of knowledge graph,an entity representation method of knowledge graph is proposed,which combines both entity attributes and relations to generate entity representation vector.Secondly,in order to solve the problem of domain offset between different knowledge graphs,the target domain entity representation and alignment methods based on domain adversarial learning are designed,which makes the target domain entity knowledge graph entity is aligned,and the target domain entity representation network is pre-trained at the same time.Thirdly,in order to improve the model performance of entity alignment in the target domain,a method based on domain adaption learning is proposed,which transfers the model and parameters pre-trained in the source domain to the target domain.Finally,the experimental results conducted on realworld public datasets show that the proposed algorithm outperforms the existing entity alignment algorithms under MRR index.Aiming at the entity alignment task in the scenario where both the source domain and the target domain have a small number of alignment labels,this thesis proposes a semi-supervised knowledge graph entity alignment algorithm based on multi-task learning.Firstly,in order to improve the quality of entity representation in knowledge graphs,an entityrelation learning method based on attention mechanism is designed,which pays attention to neighbor entities with higher relational weights.In particular,combined with graph convolutional neural networks,the structure and attributes in graphs are utilized at the same time to generate entity representations.Secondly,in order to expand alignment seeds,an iterative alignment seeds expansion method is proposed to continuously update and expand the aligned entities.Thirdly,in order to improve the performance of multiple entity alignment tasks at the same time,a collaborative entity alignment method of multi-task based on parameter sharing is designed to enhance between multiple related entity alignment tasks.Finally,the experimental results conducted on real-world datasets demonstrate that the proposed algorithm effectively improve the alignment effectiveness.