Effective Personalized Recommendation using United Auxiliary Domain based Weighted Rating Model in Large-Scale BT Download Datasets
Yue Liu, Binkai Shi -, Guobing Zou, Zhe Xu · Journal of Convergence Information Technology · 2013
In a personalized recommendation system, users can conveniently access information that they really need along with the exponential growth of different kinds of Web data. Currently, most sophisticated off-the-shelf personalized recommendation techniques are based on the idea of collaborative filtering which can help users to find their favorites and interests. However, many important research issues still exist and need to be solved in collaborative filtering for effective personalized recommendation, especially including sparse data and cold start. To address these research issues, this paper first proposes a novel transfer learning model, called UADWR (United Auxiliary Domain based Weighted Rating model) for collaborative filtering, where an united auxiliary domain is defined and acquired by cross-domain clustering, and then the transfer algorithm is applied to generate a weighted rating model so that it can be used for personalized recommendation. Extensive evaluation on large-scale online BT dataset shows that the effectiveness of the proposed method and its applicability for personalized recommendation.