The Research Of Top-N Recommendation Algorithms Based On Learning To Rank

With the coming of data era, the "information overload" problem people are being faced is beding serious increasingly, and the search engines used to help people get information can’t no more meet the personalized information needs of people for different purposes in different contexts and at different times. However, recommendation systems, as an important branch of personalization research, are able to provide people with related information from huge amounts of data, and effectively alleviate the contradiction of imbalance between information production and information access caused by "information overload" information.Currently, the study of recommendation systems attracts extensive attention of many scholars in different research fields, and they have made great progress with the complexity of recommended scene, the challenges and problems from recommendation systems still need further research and optimized solution. In this paper we mainly focus on how to improve the Top-N recommendation, solve the problems of effectiveness and enhance the trade-off between accuracy and diversity. The main contents and contributions of our work are as following:1. Transforming the recommendation problem into the sorting problem and presenting the recommendation algorithm framework based on learning to rank(LTR).LTR which helps improve the recommendation eventually is used to solve recommendation problems of mutil feature dimentions with combineing different recomemdation model effectivelyand optimziing the weights of model automatically..2. On the basis of t the recommendation algorithm framework based on LTR, we can fuse other recommendation algorithms and modelsto form hybrid recommendation model for specific scenefor improving the recommendation accuracy. In this paper, the models of LTR and ListRank-MF are fused to achive the advantages of both. The experimental results show that it improves the recommendation accuracy effectively.3. A model is presented to represent diversity features based on entropy, twhich uses the concept of entropy to represent diversity features effectively based on attributes of users and items. And the experimental results show that it gains an enhancement of trade-off between accuracy and diversity.4. In this paper, we study the feature selection algorithms of the model based on LTR. After extracting the latent features and score features, we perform feature selection, to reduce the feature dimentions, for improving the training efficiency of sorting model. The results of our experiment show that after performing feature selection, the model starts to converge after a few iterations and improve the recommendation accuracy to some extend.5. In terms of the effectiveness in recommedation systems, we analyse and summarize out some fators that affect the effectiveness of recommendation. The sorting model of recent feedback is proposed by improving the basic recommendation framework based on LTR with embeding effectiveness fators. Experimental results show that after the second sorting of users’ recent feedback and basic recommendation output, the recommendation accuracy is improved effectively. Furthermore, we also explores the application of online learning algorithm based on the sorting model of recent feedback. The simulation expriments show, the model can still output a high recommendation accuracy after being updated online in real time.