Research On Network Alignment Based On Local Structure Feature Enhancement

With the vigorous development of Internet technology,social networks naturally penetrate into our lives.Different social networks provide users with different services.For example,people will shop through Taobao and communicate with friends through Wechat.Social networks not only bring convenience to people,but also produce a large number of user data.However,the user data on a single social network often has the problems of uneven distribution,high noise,dynamic change and uneven quality.In order to depict users in an all-round and deep level and provide more available information,network alignment has become a hot issue in the current social network computing.Network alignment aims to find corresponding nodes from different networks.A famous network alignment application is to map different social network accounts belonging to the same person.However,in the real world,networks often have problems such as high noise and data sparsity,and the existing network alignment technology has not been well solved.Therefore,this paper studies these two problems and proposes a new network alignment method.Considering that the knowledge representation learning method can well alleviate the problem of data sparsity and noise in network alignment,this paper first studies the existing knowledge representation learning methods.Through the analysis of the existing models,it is found that the existing models either only describe the relation patterns,or only model the multi-fold relations,and do not consider these two aspects at the same time.To solve this problem,this paper proposes a knowledge representation learning model MRotat E based on relational rotation and entity rotation.On the one hand,MRotat E describes the relation patterns through relational rotation.On the other hand,MRotat E learns the multi-fold relations through entity rotation.On this basis,the above features are uniformly represented by scoring function.Then,in order to more comprehensively test the effect and robustness of the model,four datasets with obvious differences in the field of knowledge representation learning are used to evaluate the performance of the model in the link prediction task,and case analysis,parameter analysis and ablation experiments are carried out.The experimental results show that MRotat E model has state-of-the-art performance in link prediction,especially in the characterization of multi-fold relations.After optimizing the existing knowledge representation learning model,this paper focuses on the network alignment method based on network representation learning.The existing network alignment methods based on network representation learning utilize the scalability of graph embedding to deal with large networks,but these methods only rely on topology information and are vulnerable to structural noise,resulting in the poor generalization ability of the model.To solve this problem,this paper proposes a network alignment model based on iterative deep graph learning.Firstly,the model iteratively learns a better network structure by the alignment prediction loss function and graph regularization loss function to better match the updated network with the alignment prediction task,so as to alleviate the structural noise in the network.Then,after getting the updated network,the model aligns the two networks through the known anchor chain to complete the network alignment task.Finally,in order to evaluate the performance of the model in alignment prediction task,alignment prediction experiments are carried out on three real datasets,and as well as parameter analysis.The experimental results show that the model has outstanding performance on large-scale datasets and has strong robustness.Although the network alignment model based on iterative deep graph learning can well capture the global network structure characteristics and help alleviate the problem of network structure noise,it ignores the node noise and the data sparsity of real networks,which makes the performance of the model on sparse datasets relatively weak.To solve this problem,a network alignment model based on local structural feature enhancement is proposed in this paper.Firstly,the model learns better global network structure features through iterative deep graph learning method.Then,the knowledge representation learning model is embedded into the alignment method to better learn local structure features and alleviate the problem of data sparsity.Finally,in order to test the effectiveness and robustness of the model,the performance of sparse datasets in alignment prediction task is evaluated,and as well as parameter analysis experiments.The experimental results show that the network alignment model based on local structure feature enhancement has better alignment prediction performance on sparse datasets,especially in the case of weak supervision.In a word,aiming at the problems of network noise and data sparsity in network alignment technology,this paper proposes new knowledge representation learning model and network alignment models,and carries out sufficient experiments to prove the effectiveness and rationality of the models.The experimental results show that the model proposed in this paper can obtain better knowledge representation and integrate social networks,so as to provide guarantee for user personalized recommendation,dynamic tracking of key users and network security.