Research On Topic-based Dynamic Link Prediction Method

Recently in e-commerce sites and social media,with the development of the information techoology,a large number of personal information data burst in unstoppable posture,but it make users obtain more worthless information from e-commerce sites and social media.So link prediction of the user's personalized is coming.According to the user's degree of similarity in a subject,it is to predict the potential link between users and products.But it is mainly to predict a potential relationship for a social network in the network structure similarity according to the previous link prediction.While the node of text information content is ignored.Simultaneously,the dynamic text information with time isn't taken into account,The traditional link forecasting method isn't match to the present situation no longer.Based on the above problems,this paper studies a dynamic link prediction method based on topics.The research work and new ideas of the paper mainly include the following:Firstly,The definition of social network and link prediction and the related concepts of topic on social network are given.Further,it is to give the formal description,and also to give the definition of dynamic link prediction based on topic,it is to use the graph to represent the social network,to lay the foundation for the link prediction in the following two chapters.Then,the paper studies and implements a dynamic link prediction method based on topics(DLPMT)in social networks.DLPMT method can be studied from two parts specificly.The first section is to research a dynamic topic mining method TDFM based on the DTM model,Aiming at the link relation between the link of the users and the documents in different time slices in the social network,the topic distribution of the documents with the time evolution is excavated,and the dynamic topic distribution of the user is obtained.For the second section,the similarity of the user's interest vector is calculated according to the theme distribution of the user excavated by the dynamic topic finding method TDFM,it is to obtain the sort of the possibility of the potential link relation among the users,the most in the front relations is the most possible link relationship.DLPMT guarantees the latestity of the acquired subject attribute,thus improving the accuracy of the link prediction.Thirdly,several experiments are designed and implemented.The experiment results on Amazon and CiteSeer Citation data set improve the new proposed method of the paper's thefeasibility and correctness.