| With the rapid development of Internet and the Electronic Commerce, the Recommender Systems based on Collaborative filtering (CF) technique have gained great importance from both the academia and industry. The CF approach tries to predict the utility of items for a particular user based on the items previously rated by other users. Since users always rate few items in practice, the sparsity problem is still a great challenge to researchers and bottleneck for the recommendation systems, even though the last two decades witnessed a lot of good algorithms and models trying to alleviate that problem.The social tags, popular along with the advance of web2.0, is a new cue for the recommendation systems. On one hand, it may reflect users'preference; on the other hand, it may describe items'semantic content. In this thesis, we focus on how to make full use of the social tags to improve the recommender systems.More specifically, the tripartite graph model is used to represent the users, items, social tags and their relationships and the random walk algorithm is applied to rank the items for a particular user. We build this tripartite graph in two different ways, that is, the user-centered approach and item-centered approach by trying to explore the user-tag and the item-tag relationships respectively, to try to discuss the utility of the social tags and the experiment results show the effectiveness of our proposed algorithms and the pros and cons of them.Besides, in order to alleviate the negative effects of social tags, such like noisy and sparsity, the Lasso logistic regression models is conducted to find similar tags in their semantic space. Thus we could expend the relationship between items and tags, which means that the items could be described more accurately, by annotating items using these similar tags.At last, we review the current work on Matrix Complete, and some prior experiments have been conducted to see how to apply this model to collaborative filtering area in the future. |
PDF Full Text Request