Recommended Changes Algorithm User Interest Based Collaborative Filtering

As the Internet popularization, people can get more and more information in it.But with the growth of information in network, Internet users can’t get the reallyuseful information when they face large amounts of information. Thus theInformation efficiency of Internet users is reduced. This kind of phenomenon isso-called "Information overload" problem. Personalized Recommendation is a goodway to solve the "information overload" problem. It can recommend the most neededresources for the users by studying the different user’s interest. And it also can greatlymeet the information demand of users in the era of "big data". Collaborative filteringis one of the more successful and widely applied personalized recommendationtechnologies. It has been widely concerned by E-commerce businesses and researchscholars. And it becomes one of the most important research issues in the recommendtechnical field.Getting users’ interests and preferences timely and accurate is the foundation ofpersonalized recommendation. As time goes on, the user’s interests will change in away. The existing collaborative filtering recommendation can’t reflect the changes ofuser interest. It also can’t provide users with accurate recommendation. As a result theaccuracy of recommendation is reduced.The following works are done in this paper:The concept, principle and classification of recommendation system areanalyzed and summarized. Then the idea, principle and method of collaborativefiltering recommendation are summarized and the basic framework of it is also gave.Based on above summary, I take a deep insight into the advantages and problems ofcollaborative filtering recommendation and clarify the key problems of user interestchange based recommendations which are needed to solve.From the perspectives of user and system to elaborate the issue of user interestchange. From the aspects of psychological and behavioral to explain the causes of user interest change. Using the summary as the key to solve the problem of userinterest change for collaborative filtering recommendation system.This paper analyze that how the issue of user interest change to affect the twocore processes of collaborative filtering recommendation, which are the process ofnearest neighbor set generation and the process of user-prediction score calculation.Then from the two process to improve the accuracy of collaborative filteringrecommendation algorithm.Based on summarizing the methods and ideas of existing research this paper givethe classification of user interest change-based collaborative filteringrecommendation according to the different ways to deal with users-projects scorematrix. The classifications are score weighting, score selection and knowledge based.This paper elaborated every classification and summed up the advantages anddisadvantages of each classification.Based on the summarized, Forgetting curve is used as the theoretical basis andthe principle of weight design is gave. User access time and user access frequency areused to describe user interest change. A new weight model is proposed for measuringthe user interest change. The efficient combination of the new data weight and theitem-based collaborative filtering recommendation is used to generaterecommendations.MovieLens data set is used as the experimental data sets to compare the newalgorithm with the traditional collaborative filtering recommendation algorithm.Experimental results show that the new algorithm proposed is better than thetraditional collaborative filtering recommendation.The following three aspects are the innovation of this paper:The issue of user interest change is deeply analyzed. This paper in-depth,inductive The affecting factors of user interest change are summed up and theinfluence of user interest change for collaborative filtering recommendation issummarized. The summary provides the key idea for collaborative filteringrecommendation system to solve the issue of user interest change. This paper give the classification of user interest change-based collaborativefiltering recommendation according to the different ways to deal with users-projectsscore matrix. The classifications are score weighting, score selection and knowledgebased. This paper elaborated every classification and summed up the advantages anddisadvantages of each classification. It highlights that the score matrix of users-project is the key to solve the user interest change and get the user’s interest.Using Forgetting curve as the theoretical basis and using user access time anduser access frequency to design the weight model The efficient combination of thenew data weight and the item-based collaborative filtering recommendation is used togenerate recommendations.The research significance of this paper as follows:In theory, this research can enrich the collaborative filtering recommendation ofuser interest in exploring the theoretical preference, and provide a new idea and a newdirection for the study of user interest change based collaborative filteringrecommendation.In fact, it can let E-commerce sites to make better recommendations task to meetthe needs of users and improve loyalty of user for e-commerce sites and promotebetter development of e-commerce sites.