The Research Of Social Network-based Personalized Recommendation Systems

Accompany with the rapid development of Internet and Information Tech, and the wide spread of smart terminal, user can create and interact information at anytime and any place, which makes data on the Internet grows exponentially. In this information explosion era, a phenomenon named Information Overhead appears. To address this issue, we can adopt Information Search and Information Recommendation Tech. Information Search returns the same result from different user's demands, which can't meet users' personality, while Information Recommendation presents user their interesting personal information. For traditional recommendation systems do not have good user experience because they miss user's social relationship, this thesis takes use of the trust social relationship between users in the SNS, recommends users individual information from their close social friends, makes recommendation much more purposeful and accurate.We deploys the concept of SNS of Sociology and AFSA in AI area to individual recommendation algorithm in this thesis. We crawl users' tag information from the Web, reconstruct users' tag network using methodology of Social Networks Analysis, present models of users' tag networks, social relationship networks and their composite. After mining social relationship among users, we extend social networks effectively by novel social relationships networks expanding algorithm. After mining neighboring users' information, utilizing the mapping relationship between user tag networks and social relationship networks, we propose TF recommendation algorithm, which returns best recommendation result by adopting two-level mixed network model to calculate the similarity of tags. We propose another recommendation algorithm based on Discrete Event AFSA model, in which we construct artificial fishes to represent target user's tag information and simulate group's collective activity, finally we get the global optimal solution.Based on Sina microblog platform, we crawl real experimental datasets, which include user's profile information, tag information and social relationship information. Then, we build users' tag networks and social relationship networks. Based on these real datasets, we do extensive experiment to test the tag-based recommendation algorithm, TF algorithm and AFSA. Result shows that TF algorithm and AFSA have significant performance improvement than tag-based recommended algorithm in precision radio and recall radio. Even more, AFSA's time cost is controllable, especially when the network scale is huge. We can get significant improvement in execution speed while sacrificing a little accuracy.