Personalized Recommendation Method In Dynamic And Multidimensional Social Networks

At the present time, information transference in the age of Internet has proudly changed theway of information sharing and Web has become the main way for people to access toinformation. The emergence of the search engine satisfied the need of information retrieval in acertain degree, however, it can’t meet the users’demand from different fields and different levels.Therefore, personalized recommendation technology is born from the need of informationsearching, it is one model of personalized service and its nature is information filtering.Personalized recommendation system is not only of great value in social economic, but alsoin scientific research. The classic recommendation method is collaborative filtering meanwhilerecommendation system based of network is novel. Generally speaking, recommendationalgorithm studies users’interest in single resource network, not in multidimensional networkwhich consists of various resources. Therefore, the personalize recommendation inmultidimensional network is a up-to-date point of view.We was inspired by the thought of collaborative filtering and recommendation base onnetwork structure and studied the theory of social network and complex network , then proposeda dynamic and multidimensional social network personalized recommendation algorithm byimporting the analysis method of multidimensional network and complex network. Firstly weproposed the definition of multidimensional overlapped network and its mapping network tobuild the multidimensional weighted network model between users. Then we imported localworld evolution theory and formed the evolution regulation for the proposed algorithm, by dintof which we founded the dynamic and multidimensional network model. Next the algorithmmade use of CPM algorithm which can recognize overlapped network cluster to search neighborsand did recommendation finally.The main work of this paper is as follows:1. By analyzing the concept and feature of social network and doing plenty of research onmultidimensional network definition and users’activities regularity in such network, the paperproposed a explicit mathematical definition of multidimensional network and its overlappingnetwork. Though there had been many concepts and definitions of multidimensional network, there have not been a unified mathematical definition now. We researched the forming process ofmultidimensional network and the transformation between single network and multidimensionalnetwork , and summarized most of the available methods. Then proposed a definition ofmultidimensional overlapped network and its mapping network which is not common used butcan depict the forming process of such network clearly. We make use of such way to descript theusers in personalized recommendation, which has changed the original modeling method whichonly utilizes the user profile.2. In the users’multidimensional weighted network, the paper analyzed its complex networkfeatures, especially the local-world evolution regulation. According to the classic local-worldevolution regulation, the paper changed the formula of connection percentage by importing thesimilarity between users and proposed the local-world evolution regulation which is suitablefor the proposed personalized recommendation algorithm and contribute to the dynamic andmultidimensional network. The founded dynamic and multidimensional network model is theprerequisite of personalized recommendation and the model to excavate and refresh users’data.3. The paper made use of CPM algorithm which can recognize the overlapped networkcluster to clusters users. On one side, the dynamic and multidimensional network model we builthas the feature of complex network. On the other side, there probably exists overlay when searchneighbors because of the variety of users’interests and the feature of multidimensionaloverlapping network. Therefore, it conforms to the network environment of personalizedrecommendation in this paper to use the complex network cluster method to search neighbors. Inaddition, the proposed algorithm utilized certain manner on the basis of users’similarity tosearch the nearest neighbors and proposed the recommendation strategy finally.4. The experiment for the proposed personalized recommendation algorithm in dynamic andmultidimensional network has verified the effect of the algorithm in several aspects. Wecompared our algorithm with common recommendation methods, validated the dynamics factorand the clustering algorithm. We utilized different certain evaluation standard to demonstrate theoperation of the proposed algorithm and presented the practical recommendation system model.