Research On Collaborative Filtering Recommendation Algorithms Based On Rough Set And Trust Network

With the vigorous development of the network technology,we entered an era of information explosion.It is difficult for people to find the required information quickly in the vast amounts of information.In the face of information overload,recommended system arises at the historic moment.At present,collaborative filtering recommendation system is one of the most widely applied recommendation systems.It establishes model through analysis historic behavior and personal preferences of the users so as to refer commodities people might like to users.In fact,most users made the choice of friends they trust,so trust network is applied to the recommend system.Although the collaborative filtering recommendation system based on trust network has achieved a certain effect,there are still some deficiency and challenges in efficiency and accuracy.This paper study the improvement method focused on the following problems faced by collaborative filtering recommendation algorithm.1? Data sparse problem.In the existing recommender systems,the number of score users made is very little.With the increasing of users and the project number in recommendation system,the user-rating-data matrix is becoming more and more sparse.The traditional collaborative filtering recommendation system recommends projects on the basis of user's score,so the recommendation accuracy is not high because of the sparse user-rating-data matrix.There are large user groups in the recommendation system,user's trust data is also very limited,it can't solve the data sparse problem fundamentally by combining trust network and collaborative filtering recommendation system.This paper proposed the collaborative filtering recommendation algorithm based on rough set and trust network,using the ROUSTIDA to fill the user-rating-data matrix to decrease the sparsity of matrix,than recommend by collaborative filtering recommendation algorithm.Combining similarity and credibility of users to form a new matrix,this can make the recommendation result more accurate.2? New user problem.It is difficult to calculate their similarity for users who don't have score or have little score in recommendation system,so the result of collaborative filtering recommendation algorithm isn't accurate.This paper introduced the expert users evaluation pattern,chose some expert users from recommendation system according to liveness,credibility and specialty,established expert user-rating-data matrix.We think the recommend of expert users are more reliable,so only use expert user-rating-data matrix to recommend for the new users.In order to let ROUSTIDA algorithm fit the extremely sparse matrix,this paper improved ROUSTIDA algorithm.When calculating the extended discernibility matrix,we use the method of difference instead of equivalence,broaden the condition of similar score,two users are more similar if the difference is smaller.Experiments show that the collaborative filtering recommendation system based on rough set and trust network produced in this paper can solve the data sparse problem and the cold-start problem,it works with higher efficiency and accuracy,and its performance of recommendation improved.