Research On Social Relationship Based On Topic Model In Review Network

In recent years,with the rapid development of Internet technology and the emergence of various new social platforms,social network-related research has been promoted in the field of data mining and analysis.Among them,network community structure and social relations,as the focus of social network research,have also received widespread attention.Community discovery can dig out the similar relationship between nodes in social network,so as to carry out further data analysis and application.The development of location-based service technology provides the premise for the emergence of location-based review network.This new type of social network not only contains the topological information of the social relations in the traditional social network,but also includes the information that users check in at different locations and the information of geographical dimension.Because of the particularity and complexity of its network structure,it presents new challenges to the analysis and research of the traditional social relations.To solve this problem,this thesis aims at the social topology structure and implicit semantic information based on geographical location in the review network scenario.A Multidimensional social relationship analysis model(MSRAM)was designed based on two aspects of dynamic friend interaction and user check-in behavior.Therefore,the results of the divided communities can meet the requirements of tight topology connection and high similarity of behavior patterns among users in the community in terms of social relations,geographical distribution and interest subject index.Based on the parameter transition probability formula and joint probability distribution expression of the model,a Gibbs sampling algorithm for solving the model is proposed according to the sampling rules of hidden variables and parameter iteration rules.To solve the problem of network data sparsity,this paper proposes a data dimension equalization algorithm based on Biterm theme model,which makes the output distribution of the model smoother.To verify the effectiveness of the proposed model,Yelp reviews network public data sets selected in the study as a real data source,after data preprocessing by experiment data,the output of the above model in the user belongs to probability distribution of community structure,verified the model on the internal similarity and the module has the ideal results,It can realize the precise community division of the whole experimental data network.By comparing with several other existing clustering algorithms of social networks,it is illustrated that the model proposed in this thesis has the advantages of closer relationship between nodes within the community and higher similarity,and alleviates the problem of data sparsity to a certain extent.