Collaborative Filtering Recommendation Algorithm Based On Trust Relationship

With Web2.0 and mobile intermet's rapid rising,the way of getting information has changed greatly,and informatization has been deepened day by day.Information from internet grows so rapidly,we have entered an age of Big Data.It has been more and more difficult for customers to find truly interesting and valuable information from massive data,and it also makes a large amount of message nobody cares become long-tail,which leads to the reduction of information utilization.Propelled by what is said above,recommendation system comes into being.By analyzing historical behavior,it can forecast consumers5 interests and push notification actively through the related means of Data Mining technologies.At present,recommendation system has been widely used into many fields.In addition,social network is increasingly developed in recent years,and it connects people in the world through the Internet.It can largely make up the shortcoming of the traditional recommendation algorithms by applying information people exchanged trough Internet into recommendation process.This paper fully explores the relationship between the social network users,and combines with the traditional algorithm to predict item ratings,so that we can improve the quality and reliability of the recommendation.The main research work is as follows:(1)Under the condition that the user rating data are extremely sparse,traditional calculation method of user similarity has some shortcomings,resulting in a dramatic decrease in the recommended quality.This paper puts forwards the concept of local similarity,introducing overlap factor to revise the concept,and then obtain global similarity by adding all revised local similarities.The improved user similarity algorithm can avoid unreasonable phenomenon that the common rating data is scarce but user similarity is high.(2)The traditional collaborative filtering algorithm takes the similarity between the users as the final weight,and then generates neighbor users,without considering the influence of trust relationship between users.This paper argues that the trust relationship between users can also be used as a recommended weight to generate neighbors.About data sparseness in collaborative filtering algorithm,this paper presents a trust relation transfer rule,which to a certain extent eases the problem above.(3)Traditional collaborative filtering algorithm recommend only relies on user-item rating data,without considering the expiring of information that the earlier the score is,the lower the value.In order to predict more accurately,this paper introduces improved time function in predicting rating phase to improve recommendation quality.At last,in order to verify the validity of improved algorithm,the simulation experiments will be taken by using MovieLens,The experiment results show that the proposed algorithm has higher accuracy than traditional collaborative filtering algorithm.