A trust-based Top-K recommender system using social tagging network

With the expansion of e-commerce, recommender systems are drawing more and more attentions. Collaborative Filtering(CF) is the most popular algorithm used for recommendation, but it performs not very well for sparse data and new users. The emergence of trust-based recommender system has solved the problem of CF in a better way. A system of this kind is commonly constructed based upon social network with trust relations. It contains not only user-item rating relations, but also friendships between users, and the friendships are very meaningful in recommendation. However, the trust values in the networks are all specified by users, which are subjective processes. In this paper, we propose a Top-K recommender system on social tagging network, and design a user-item rating matrix construction method on user browsed or searched information. We use tags, which can be regarded as users and items feature information, to compute the similarity between users or items. Moreover, we propose a Top-K recommender system construction method on the network with trust values computed from users´ interest similarity. In our experiments we use the Last fm dataset, and we employ the RMSE and hit-radios benchmarks to evaluate the quality and performance of prediction on single item and Top-K recommendation. We compare our approach with two traditional CF algorithms. The experimental results show that our system has good performance, and it solves the defects of CF and existing Trust-based recommender systems.