Research And Application Of Data Sparseness Problem In Collaborative Filtering Recommenderation

Personalized recommendation technology is an effective solution for Internet business to propose resources to users who will be interested in.By studying the user's interest preferences,personalized recommendation technique recommends the resources that most likely to be in line with the user's interests actively and timely.helping users to locate the resources they need accurately and quickly.The personalized recommendation technology based on collaborative filtering algorithm is the one of the most successful and widely used technologies and it has been successfully applied to various areas of personalized recommendations.However,basing on collaborative filtering algorithm to construct the personalized recommendation system is also facing many problems,such as cold start,the user evaluation of malicious fraud,scalability and high performance.What is worse,the evaluation data sparseness problem caused by the uncertainty of user's feedback and other factors,would not only severely reduce the accuracy of recommendation,but also have a severe influence on the building of collaborative filtering-based personalized recommendation system as a whole,the performance and the scalability of the system.In this paper,depending on the research and the learning of personalized recommendation technology as well as the collaborative filtering algorithm,aiming at the sparsity problem of user data we propose a hybrid weighted prediction algorithm.According to the characters and the popularity of the visited data,the algorithm predicts the visited but not evaluated data and fills them in the evaluation matrix to reduce the sparsity of the evaluation matrix caused by the the uncertainty of user 's feedback and improve the accuracy of recommendation.At the same time,because the algorithm does not conflict with the online recommendation,it can be identified and caculated offline which provides a good scalability.Experimental results on data sets in MoiveLense show that compared to the personalized recommendations based on the traditional collaborative filtering algorithm,the proposed algorithm can significantly reduce the sparsity of the evaluation data,increase the accuracy and the scalability of the recommendation.As the proposed algorithm can be directly used in the personalized recommendation system based on the collaborative filtering algorithm,to predict and fill the user's feedback data,this research is meaningful...