The Research And Implementation Of Dynamic Social Network User Alignment

In recent years,with the birth and popularization of more and more social platforms,people are increasingly active in various virtual social networking sites.Because different social networks will show different functions,users will show different activities on various websites,and their behavioral data will have different emphasis.In order to make full use of social network data and integrate the user information scattered in each network,a very important task is to align and match the user's social accounts in multiple networks so as to connect multiple social networks,and the corresponding problem is called user alignment problem.Cross social network user alignment provides a feasible solution for the integration of multi-source data,which can help many social network applications and facilitate the exploration and research of social networks.In this thesis,a dynamic social network user alignment algorithm based on network embedding representation is proposed,which uses the structural and attributed characteristics of dynamic social network to achieve accurate user alignment.Dynamics is an important property of social networks,but this important dimension has not been effectively used in social network alignment.Structural and attributed features are two important features of social networks,which provide complementary and comprehensive social network information.In this thesis,we first mine the structural and attributed characteristics of dynamic social network,then use deep autoencoder,recurrent neural network and attention mechanism to capture the dynamic behavioral patterns of social network users,and use the feature fusion method of maximum likelihood estimation to describe social network users more comprehensively and accurately.Finally,we use the method of measurement learning and subspace learning to achieve user alignment of dynamic social networks in a semi-supervised way.The whole algorithm framework realizes end-to-end joint optimizationThe experimental results show that the dynamic user alignment algorithm designed in this thesis can effectively utilize the dynamics of social networks,and effectively integrate the structural and attributed characteristics of social networks,so as to achieve better user alignment effect than the state-of-the-art alignment methods.