Research On Community Detection And Collaborative Filtering Recommendation Technology In Social Networks

With the rapid development of Internet, various social media have sprung up, which provide different ways for people’s communication and interaction. Since the accumulated interaction data have great academic value and broad application prospect, how to analyze the social network and mining its hidden rules has become the focus of public attention.Both the community detection and collaborative filtering recommendation are hot topics in the field of social network analysis. Community detection technology is helpful to reveal the networks’inherent structures and thoroughly analyze the complicated relationships. On the other hand, social network analysis boosts the development of recommendations and precision marketing; the social network fused recommendation can also alleviate the data sparsity problem. In this paper, we’ll focus on the detection of meaningful communities, and how to improve the performance of recommender systems with the user relations. In conclusion, the main contributions of this dissertation are as follows:1. The existing community detection algorithms mainly focused on the process of the nodes’cluster, while the nodes’similarities are also a critical factor affecting the performance of algorithms. This paper proposes a novel community detection framework based on iterative affinity refining. By iteratively modifying the affinity of node pairs with a self-taught approach, the refined similarity matrix can depict the relationships between node pairs more precisely, which will lead to the performance improvement of community detection.2. For the community detection of the multi-relational networks, we have introduced a novel framework named SLS, which mines the hidden community structure of multi-relational network by a two-stage process:firstly we extract the shared local structure based on selected partition results, a part of the nodes will be divided into several groups which can be regarded as the subset of the community; then based on the existing partitioned clusters we can predict cluster assignments of the remaining nodes. The proposed model has been proved to be robust and superior by comparative tests on both synthetic and real networks.3. Aiming at the problem of data sparsity for collaborative filtering, a social network based recommendation model is proposed. It can not only capture the social homogeneity factor, where users’behavior can be predicted by his/her friends in the social network, but also learn the heterogeneity factor which indicate the users’ individual taste. In the end, the comparison experiments on several real datasets validated the superior of our proposed model.