Research On Personalized Recommendation System Algorithm Based On Collaborative Filtering

With the rapid development of Internet technology and the advent of the era of big data,the network information resources have rapidly expanded and formed massive information.When people enjoy the convenience of large data age,they also face with the problem that it is increasingly difficult to find information that people need from the huge information resources,that is the phenomenon named ‘information overload'.In order to solve this problem,many scholars have put forward a solution based on personalized recommendation technology,which can effectively alleviate the ‘information overload'.Among personalized recommendation technology,collaborative filtering algorithm is the most widely used technology.This paper will take the collaborative filtering algorithm as the research object,the problems such as accuracy,expansibility and data sparsity will be studied and the corresponding improved algorithm will be proposed in this paper.The research work of this paper is as follows:1.Overview of recommendation system and recommendation algorithm.First,this paper introduces the basic concept of recommendation system in detail;Then,we analyze several recommended algorithms commonly used in recommendation system;Finally,the paper focuses on the classification of collaborative filtering algorithm and the implementation steps of userbased collaborative filtering algorithm.2.Research on improved collaborative filtering algorithm with fusing item attribute and interest change.In order to alleviate the problems of finding the nearest neighbors inaccurately and ignoring the change of user interest among the traditional collaborative filtering algorithm,this paper presents an improved collaborative filtering algorithm which integrates item attributes and interest changes.First,for the problem of finding the nearest neighbors inaccurately,based on the user-item scoring matrix,this paper constructs the user preference matrix which can reflect the user's preference information of item attributes.Based on the user preference matrix,the similarity of user to the item preference information is calculated,the similarity is weighted to the traditional user similarity based on the scoring matrix.Then,an improved method of comprehensive similarity calculation is obtained.Next,for the problem of ignoring the change of user interest,this paper adds a decay function that reflects the change of user interest during the scoring prediction phase.The experimental results prove that the improved collaborative filtering algorithm which integrates item attributes and interest changes through the improvement of similarity and scoring prediction formulas can effectively alleviate the problems of finding the nearest neighbors inaccurately and ignoring the change of user interest faced by the traditional collaborative filtering algorithm and improve the accuracy of recommendation result.3.Research on improved collaborative filtering algorithm based on fuzzy C-means and matrix completion.First,aiming at the scalability problem of traditional collaborative filtering algorithm,this paper proposes an improved clustering-based collaborative filtering method.It only finds the nearest neighbors in the clusters,this method greatly reduces the search space and improves the recommendation efficiency.At the same time,considering the problem that the clustering effect of traditional k-means algorithm is not ideal,when clustering users and items,this paper uses fuzzy clustering to replace the traditional k-means clustering algorithm in order to improve the accuracy of the nearest neighbors;Then,in order to alleviate the problem of data sparsity faced by traditional collaborative filtering algorithm,in this paper,a method of filling sparse matrices with weighted Slope One algorithm is proposed based on item clustering.When the sparse matrix is filled with the weighted Slope One algorithm,considering the phenomenon of inaccurate filling when the number of common scoring user is less,this paper presents a method to improve the quality of the data being filled by setting threshold factor for Slope One algorithm;Finally,on the basis of the above improvements,an improved collaborative filtering based on fuzzy C-means and matrix completion is proposed.The improved collaborative filtering algorithm is clustered simultaneously in the user and item direction.First,it uses the user preference matrix and the original scoring matrix to cluster the user and the item respectively;Then,the original matrix is filled with the improved Slope One algorithm on the basis of the item clustering.The user similarity is obtained by using the improved similarity formula based on filled matrix;Finally,the user-based collaborative filtering algorithm is used to recommend.The experimental results show that the improved collaborative filtering based on fuzzy C-means and matrix completion can effectively mitigate the impact of data sparsity while ensuring real-time performance,and significantly improve the accuracy of the recommendation result.