The Research Of Ranking Recommendation Based On Implicit Feedback

With the rapid development of the Internet, we are in an era of information overload. The traditional search engine has not been able to satisfy the people's increasing personalized demand of information. In order to help people to effectively filter the great deal of information, the recommender systems emerge as the times require. Nowadays, most of the recommender algorithms are based on explicit feedback data.However, a large number of implicit feedback data have not been underused. In the scenario of making Top-N recommendations, the idea of traditional recommender algorithms is based on the users' rating prediction on items, but neglects the ranking relationships between items.Therefore, in view of the above two questions, we focus on the research of ranking recommendation based on implicit feedback, the main contents are as follows:1. Make the analysis and comparison of nowadays recommender algorithms based on implicit feedback and learning to rank, and then present the improvement of related models and algorithms.2. To aim at the problem of the existing user model, which has sparse explicit attributes, this paper propose a method for mining implicit attributes of users and a hybrid user model which is based on the explicit and implicit attributes. Firstly, we construct the user's description document according to users' implicit feedback data and the social tags of the items, and then extract the user's implicit attribute through the LDA topic model. Secondly, the vector space model based on the explicit attribute and the topic model based on the implicit attribute are merged.The hybrid user model is established by the combination of the user custom modeling and the automatic modeling. Based on this, a similarity calculation method between users is proposed.3. A personalized model for transforming users' implicit feedback data into interest level is proposed, so that the ListRank-MF algorithm can be applied to the implicit feedback data. And also, the users' nearest neighbor information which is based on the hybrid user model is incorporated into the algorithm, so that the algorithm considers not only the items' sorting information of individual users, but also the mutual influence between nearest neighbors.4. We make experimental verification for the proposed hybrid user model which is based on the implicit and explicit attributes,and the UNListRank-MF algorithm -based on implicit feedback data. The experimental results show that the hybrid user model, which combines the explicit attribute and the implicit attribute, and the UNListRank-MF algorithm, can improve the performance of the recommendation effectively.