Research On Community Division And Application In Dynamic Multidimensional Networks

In recent years,information technology is developing fast and the internet is becoming moreand more popular. The social websites are increasing and more and more users share andcommunicate various information through websites. In social networks, social network analysisis, based on some principles, to conduct community division among group relational modelsbuilt between actors to make the users obtain their most interested or most popular user orinformation in the objective networks, which avails users to conduct information sharing andcommunicating faster and more directly.The traditional social network model is mainly simple static single relational model, whileto most networks, their topological structure will change obviously with the passage of time. Thetraditional method concerns about the overall features of the network. Therefore, the emergencyof a individual or community at a particular point is often omitted. At present most of themethods are to excavate community in the single relational social network, but the result cannotfully meet the needs of users. Whereas most of the realistic social networks are muti-relationnetworks. How to deal with these complex structures and get useful network information is anew challenge to the traditional social network analysis method.Based on the above problems and combined with the present literatures, this thesis makes arelevant research on the static, dynamic and multidimensional features of social networks on thebasis of complex network theory and relevant techniques.By analyzing the evolution of community, put forward the matching algorithm to obtain thestable community assembly correspondent with the actual community structures; Use thedimension reduction method presented in this paper for reducing the dimensionality of themultidimensional network, and combining with the dynamic feature of networks, to dividecommunity for constructing the dynamic multi-dimensional network. This thesis mainly includesthe following three aspects:1. Because static analysis does not consider the time of interaction, it missed the chance ofcatching the evolution model of dynamic network. Combined with the concepts and features ofsocial network, and through the analysis of the dynamic evolution process of the network, to beexactly, the testing of community evolution and the changing of community structure with thepassage of time can help us understand the potential behavior of network, this thesis proposes aframework to build model and detect the evolution of community in social networks. In thisframework, I define a series of important events for every community and formalize evolutionevent model according to its features. Use the matching algorithm to calculate the similaritybetween communities during the process of evolution to get the evolution way and effectivelyidentify and trace the similar community which is changing with time. Finally I get the stablecommunity assembly correspondent with the actual community structures. I also define theconcept of metacommunity, a series of similar community detecting with matching algorithmwhich is caught and put forward at different time periods. This thesis does not focus on individual evolution but observes the evolution of community macroscopically. Throughexperiments with the three datasets, test and verify its validity and possibility.2. Make an explanation of the multidimensional network and its dimension at some pointsaccording to the the characteristics of social network’s dynamic changes and the bipartite graphnetwork theory and also illustrate the process of dimension reduction network formed betweenusers.3. In user’s network structure model, analyze the behavioral characteristics existed in eachuser and the dynamic changes of relations between users and propose a method for calculatingthe attribute similarity, which can be used to calculate the similarity of dynamic attributesbetween users, and apply this method to recommendation system. Besides, This article usesnearest neighbor search method which is based on user similarity to give appropriaterecommendations. Experiment results show that the application of algorithm of this paper to therecommendation system is feasible and the accuracy rate of recommending is greatly improved.