The Research On Learning To Rank Via Integrating Multiple Feedbacks

With the surge of internet economy, users are gradually drown in the flood of massive information. Under this information explosion background, in order to help users efficient and accurate access to the information of their interest, recommender system came into being.Recently, recommender system has attracted a lot of attentions, which helps users to find items of interest through utilizing the user-item interaction information and/or content information associated with users and items.The interaction information (i.e., feedback) between users and items are widely exploited to build recommendation models. The feedback data in recommender systems usually comes in the form of both explicit feedback (e.g., rating) and implicit feedback (e.g., browsing histories,click logs). Although existing works have begun to utilize either explicit or implicit feedback for better recommendation, they did not make best use of these feedback information together.In this paper, we focus on study the learning to rank algorithms integrating multiple feedbacks in the recommendation scenario. We first study the personalized ranking recommendation problem by integrating multiple feedbacks, i.e., one type of explicit feedback and multiple types of implicit feedbacks. Then we propose a unified and flexible personalized ranking framework MFPR to integrate multiple feedbacks.Moreover, as there are no readily available training data, an explicit feedback based training data generation algorithm is designed to generate item pairs with more accurate partial order consistent with the multiple feedbacks for the proposed ranking model.Extensive experiments on two real-world datasets validate the effectiveness and superiority of the IPPE algorithm , the SFRP and MFPR models. In the last part, we design and develop a prototype recommender system integrating multiple feedbacks. The analysis of the recommended cases indicates that the integration of multiple feedbacks can make up better complementary information and significantly improve the recommendation performance.