Research On Hybrid Collaborative Filtering Recommendation Algorithm Based On Split-step

With the rapid increase in the amount of data in the information age,the way we get information has experienced the evolution from the portal,the search engine,to the recommendation system.In contrast,the recommendation system is more suitable for the scene that user needs are vague and not clear,and the user is passive access to information.In recent years,the recommendation system has made great progress in the field of research and application.It plays a more and more important role in improving user activity and sales performance.The recommendation system consists of two research directions,Rating prediction and Top-n Recommendation.In the early stage of the research in recommendation system,academic research mainly concentrated in the Rating Prediction.But with people’s further study people found that the Top-n Recommendation has more practical application value.Users are more concerned about whether the items recommended by the recommendation system could meet their needs.They don’t care about the accuracy of the rating predicted by the recommendation system.Although the recommendation system has developed rapidly in the past decades,the system is faced with the problem of data sparsity and cold start with the growth of data.The current analysis of the data for recommendation system focuses mainly on the rating information,ignoring the significance of the rating behavior to the user’s interest,which is important as well.There is still a possibility of further improvement of the criteria for evaluating the accuracy of the recommendation system.In this paper,we will mainly study the optimization of the Top-n recommendation problem,and put forward the solution to the above problems.The main work are as follows.(1)For data sparsity problem,a new data structure is adpted to represent sparse matrix,save storage space,and optimize the matrix operation process.(2)A "Split-step" model of user behavior and a hybrid collaborative filtering recommendation algorithm based on "Split-step" are proposed.The rating behavior is divided into two steps: Selecting Items and Rating Items.The algorithm consists of KNN hybrid collaborative filtering based on Split-step(KS),Model-based hybrid collaborative filtering based on Split-step(MS),KNN and Model-based hybrid collaborative filtering based on Split-step(KMS).(3)On the basis of NDCG,the paper puts forward a macro index NDCG+,which is able to represent the overall accuracy of the algorithm and more applicable.(4)Optimize the algorithm based on “Split-step”,including improvement of similarity computation algorithm,Weight-added KNN hybrid collaborative filtering based on Split-step(WKS).The optimization results are verified by experiments...