Research Of Social Relationship Prediction Across Networks Based On Asymmetric Neural Network

Research on user relationships in social networks will be helpful to build a more complete social platform.But current researches focus more on social relationship predictions in single social network.Due to the lack of interaction and relationship label information between users in real social networks,social relationship prediction across networks will become more and more important.Meanwhile,due to the social networks are constantly evolving,the problem of link prediction in dynamic social networks and the prediction of dynamic social relationship types across social networks are becoming hot research topics.Based on the discussion,a new asymmetric BP neural network model is proposed in this thesis.The transfer learning method is used to predict the types of social relationships across social networks.At the same time,research on link prediction and the prediction of dynamic social relationships are conducted in social networks.The specific research contents are as follows:1.The structural characteristics of social networks are analyzed in this thesis,the common features between the source social network and the target social network are grasped,and the unique characteristics of the target social network are mined.Based on the above features of the two social networks,an asymmetric neural network model based on transfer learning was proposed to predict the types of social relationships in the target social network.Through conducting experimental verification on six real online social networks and comparing with some existed methods,the proposed model can effectively predict the types of social relationships for the target social network.2.The dynamic nature of user's behavior causes constantly change in social relationships.In order to study the evolution of social relationships in dynamic social networks,the evolution law of triad motifs in dynamic directed networks is fully analyzed,and the appropriate time series analysis methods are used to predict the transition probability of triad motifs.At the same time,combining with the tightness of the connection between the internal nodes of the motifs,a link prediction model based on the evolution of the motifs and the tightness of the connection is proposed.The results show that the proposed model can achieve better link prediction results.3.Since the social relationships in social networks are constantly changing,the above classification models may not be able to predict new adding samples.Based on the above researches,the sliding window method is used to predict the type of dynamic social relationships across social networks.At last,an asymmetric neural network model based on move window is proposed for dynamic social relationships type prediction across social networks.