Research On Collaborative Filtering Recommendation Technology From The Perspective Of Community Detection

With the popularization of Internet, the quantity of web information grows with thetrend of index, and users are difficult to find the useful information from the ocean ofinformation because of information explosion. To provide users personalized informationservice, personalized recommendation systems came into being. The application ofcollaborative filtering recommendation technology is most successful and widely, butcollaborative filtering recommendation has some problem, such as sparsity, cold start,expanding and real-time recommendation, due to the limitation of recommended principle.These problems influence the accuracy and efficiency of collaborative filtering, even affectthe development of recommendation systems of the collaborative filtering. This paperintroduces the theory of community detection in the complex networks, and organicallycombined the algorithm of community detection with the algorithm of collaborativefiltering. The binary network based on users and items is mapped into user network, andthe user network is divided into several communities, and then the application ofcollaborative filtering algorithm is based on the communities. In this paper, collaborativefiltering algorithm based on community detection can solve the problems of new users andexpanding, can greatly enhance the efficiency of recommendation and meet the needs ofreal-time recommendation. The main research content is as follows.Firstly, improve random walk algorithm of community detection and mainlyreconstruct the node distance and community distance. Transition probability of t-step isreplaced by effective transition probabilities. Effective transition probabilities can avoidthe effects that the different random step values make node distance and communitydistance change，to enhance the algorithm’s robustness and stability.Secondly, propose the multi-label propagation overlapping community detectionalgorithm. Label propagation algorithm extends each node to own kinds of tags, retains allupdated labels of the node and stores them, to make each node update the labels’memory space. When the algorithm ends, the based-label post-processing obtains overlappingcommunity structure.Lastly, propose collaborative filtering algorithm based on community discovery.Combined the improved random walk community discovery algorithm and the multi-labelpropagation overlapping community detection algorithm with the based-Pearson similaritycollaborative filtering algorithm respectively.It uses MovieLens dataset as experimental data. The results show that therecommendation efficiency of collaborative filtering algorithm based on communitydetection are better than collaborative filtering algorithm based on Pearson similarity, itdecreases slightly in recommendation accuracy.