Research On Personalized Recommendation Based On Collaborative Filtering

With the rapid development of the Internet,people can easily get a lot of information in work,study and daily life.But because of the exponential growth of information resources,it is difficult for people to get valuable information from these huge information resources.That is,information resources are in a state of overload.In oreder to solve this problem,personalized recommendation technology emerges as the times require.This technology is mainly through some algorithm to analyze users' historical behavior track on the website,accurately position interests of users.Finally,it automatically recommends some interesting items to the user.In the field of personalized recommendation,collaborative filtering algorithms are widely used and many scholars are doing deep research.The paper improves the traditionnal collaborative filtering algorithm in terms of extendibility,Salton similarity lack of user's emotional attitude and user interest migration.The main researching content is as follows:(1)To solve the extendibility problem,this paper uses firstly spectral clustering algorithm to cluster the users.In this way,the target user searches for the nearest neighbor only in his cluster instead of the global scan,which greatly improves the timeliness of the recommendation.Furthermore,the spectral clustering is realized by constructing the Laplacian matrix by the user vector.The essence of this convert the features of the original user to a more easy feature vector.In this way,the problem of data sparsity can be overcome well and the effect of recommendation is better.(2)To solve Salton similarity lack of user's emotional attitude,by analyzing the user's rating and evaluation of the film,this paper draws the mapping relationship between the score and the user's emotional attitude.On this basis,the decision tree trategy is used to find the mapping relationship between the user's emotional attitude and the weight of the common present value.The weight of the common present value apply to traditional Salton similarty that the emotional attitude of user apply to traditional Salton similarty.The accuracy of the algorithm to search the nearest neighbor is greatly improved.(3)In view of the user interest migration problem,inspired by the Ebbinghaus forgetting law,a time weighted prediction model is proposed.The model can make the weight of hobby greater that is closer to the current time of the target user,and make the weight of hobby smaller that is farther to the current time of the target user.This can greatly limit the impact of the early hobby of the user on the recommendation effect and improve the intelligence level of the recommendation system.Finally,experiment on MovieLens datasets of 100 K is done and results show that the improved user based collaborative filtering algorithm has a better recommendation effect than the traditional algorithm.