Research On The Application Of Personalized Recommendation Based On Collaborative Filtering Algorithm

Nowadays,personalized recommendation system has been widely used.Recommendation algorithm is the core of the recommendation system.However,the most widely used collaborative filtering algorithm has a lot of problems.For that reason,the paper mainly consists of the following parts:First of all,in order to deal with the traditional collaborative filtering problem of poor real-time,the paper uses the clustering algorithm to reconstruct the score matrix.However,the traditional k-means clustering method,which is the method of randomly selecting clustering centers,will easily to result in the instability of clustering results.This paper adopts a method which can select the relatively stable initial cluster center to improve the traditional clustering method.By using clustering method,the users with high similarity can be divided into the same cluster.When the users search for the nearest neighbor,the target users only need to consider the class clusters with high similarity.This method speeds up the speed for searching the neighbors.To some certain extend,the real-time problem and the calculation speed of the traditional algorithm can be improved.Second,this algorithm is different from the traditional clustering which is based on collaborative filtering algorithm in the whole score matrix clustering.This paper uses latent semantic model to decompose the rating matrix.In order to prevent overfitting and achieve the original score matrix decomposition,this paper will apply the decreasing gradient algorithm to update them.After decomposing the rating matrix,there will be two low dimensional matrix.One is the user class matrix and the other one is the item matrix.The decomposed matrix is used to cluster.After decomposition,a lot of useless data are discarded,so the dimension of the cluster is reduced.At last,this paper achieves a movie recommendation system.Users can enter the user's ID number and the number of films to be recommended in the system.Based on the analysis of historical information from users,the recommendation module can find out the movies which the users are interested in.According to the predicted user's preference for the film,the recommended movies will be sorted.The system tacitly approves that the higher the prediction is,the more likely the users like the movies.According to the system recommendation results,users can choose the films they are interested in,which can save a lot of time.In order to prove the validity of the algorithm,this article has carried on theexperiment.The results show that it improves the speed and quality to a certain extent.