Research On Cross-network Entity Alignment Based On Multi-source Interaction Fusion

With the popularization of intelligent devices and the development of mobile Internet,massive user behavior information is generated on the Internet.At the same time,with the rise of various platforms,a user entity may register accounts on multiple platforms at the same time.The linkage between platforms is enhanced,and network data presents cross-platform characteristics.The analysis and mining limited to a single network under the massive data scenario can no longer meet the current practical application needs.It is of great significance for cross-network recommendation system,cross-network crime tracking and cross-language translation to dig out people and objects in different networks corresponding to the same real scene from massive network nodes through network entity alignment technology.On the one hand,the large node number,topological structure,diversity and imbalance of the network structure data make it difficult for the existing network representation learning methods to learn and extract efficient feature information.On the other hand,the virtuality of the network and the cross-domain characteristics of different networks make it difficult to obtain a large number of cross-network marking training data in real scenarios,which severely limits the practical application ability of existing supervised algorithms and models.Therefore,this paper will study from the two aspects of network representation learning and entity alignment.Firstly,this paper studies network data feature extraction and representation learning,starting from the diversity of interactions between nodes in heterogeneous networks and the hierarchical nature of neighborhood nodes,conducts an in-depth analysis of current network representation learning methods,and proposes a multi-source interactive fusion network representation learning mode l.The node neighborhood is divided into first-order neighborhood and higher-order neighborhood.The first-order neighborhood information aggregation algorithm and the higher-order neighborhood information aggregation algorithm proposed in this paper are used for information aggregation,and then the multi-order neighborhood information aggregation algorithm proposed in this paper is used for information fusion.The method has been verified on a variety of downstream tasks by using real open-source data,and the experimental results show that the proposed method has improved on various indicators to different degrees compared with the existing method.Secondly,based on the research in the aspect of network representation learning,this paper makes an in-depth analysis of the current network entity alignment method.Starting from the fact that it is difficult to obtain a large number of cross-network entity alignment marking data under the actual scenario,a residual cycle generative adversarial networks entity alignment is proposed.The model of residual generative adversarial networks alignment proposed in this paper is used to realize entity alignment learning in unsupervised scenarios.On this basis,the reconstruction loss model and cycle model are used to optimize it.Based on the real data and generated data,the proposed method is verified.The experimental results show that the method proposed in this paper has great improvement compared with the current method,and unsupervised learning also improves the applicat ion ability of the model in practical scenarios.Finally,based on the research results in the aspects of network representation learning and network entity alignment,this paper designs and implements a cross-network entity alignment prototype system according to the requirements of practical application scenarios,which verifies the feasibility of the research content in this paper from the practical application.