Research On Collaborative Filtering Algorithms Based On Heterogeneous User Feedback

With the rapid development of internet, especially the the technique of Web 2.0, we are exposed to "information explosion". To solve this problem, the technology of recommendation system is developed, in which collaborative filtering is most common used technique. In this thesis, the problem, combination of explicit feedback and implicit feedback, and the problem, combination of multiple implicit feedback, are being studied. The algorithm, Compressed Knowledge Transfer via Factorization Machine, and the algorithm, Adaptive Bayesian Personalized Ranking for Heterogeneous Implicit Feedback, are presented for the two problem respectively. The major works of research are as follows:Firstly, the algorithm, Compressed Knowledge Transfer via Factorization Machine, is proposed for the problem, combination of explicit feedback and implicit feedback. Compared to the original Factorization Machine, this algorithm can improve effectiveness of recommendation and efficiency of time.Secondly, the algorithm, Adaptive Bayesian Personalized Ranking for Heterogeneous Implicit Feedback, is proposed for the problem, combination of multiple implicit feedback. This algorithm take advantage of transfer learning to solve the problem of ’uncertainty’, which the original Bayesian Personalized Ranking cannot solve.Thirdly, experiments are designed to verify these algorithms and results show that these algorithms proposed in this paper can perform as expected.