Research On Sparsity And Scalability Problem In Collaborative Filtering

In recent years,with the rapid development of service computing and cloud computing,more and more applications in the form of cloud services provided by the developer to the user,which led to the explosive growth in the number of services.In the face of a large number of candidate service groups,how to quickly locate the service that meet the individual needs of users has b ecome one of the urgent problems to be solved.Personalized recommendation system can change the user's potential,fuzzy needs into realistic,clear needs,to help users filter invalid information,effectively alleviate this problem.Among them,collaborative filtering recommendation technology became the most successful technology of personalized service recommendation field because of its' simple process and high accuracy.However,with the increase of the number of users and services in the system,the user-item rating matrix becomes more and more sparse,which seriously affects the recommended quality of collaborative filtering algo rithm.In addition,in the face of massive data in the system,the scalability of collaborative filtering algorithms is also facing severe challenges.In this paper,we study the sparsity and scalability problem of collaborative filtering algorithm.First,Logistic function is used to calculate the user's interest based on the nonlinear relationship between service invo cations times and user interest,the user's rating data is not needed so that improve the sparseness of the algorithm.Secondly,the off-line user clustering algorithm is used to cluster the original large-scale user data.In the real-time calculation process,the nearest neighbor is found only in the cluster with the target user,and the recommendation efficiency is improved.Specifically:Firstly,a collaborative filtering recommendation algorithm based on Logistic function is proposed to solve the sparseness problem of collaborative filtering algorithm.First of all,by analyzing the user's history invoke record of the service,the invoke time is used to replace the rating data in the traditional method.Then the Logistic function is introduced to standardize the invoke times to reflect the user's interest more rationally.Finally,predicts the service that the user is interested in based on traditional collaborative filtering algorithm.Secondly,in view of the scalability issues of collaborative filtering,a collaborative filtering algorithm based on user clustering is proposed in this paper.This method first combines invoke times and service keywords to get the user's preference for the service keywords,and uses the TF-IDF method to calculate the user's preference,construct the user-keyword preference vector;then introduce the Logistic function to calculate the user's interest in the service;finally,the nearest neighbor is found only in the cluster with the target user,and the target user's interest in the target service is predicted according to the interest of the nearest neighbor to the target service.Finally,the paper analyz es and validates the effectiveness of the above methods.