A Novel Clustering Based Collaborative Filtering Recommendation System Algorithm
Traditional collaborative filtering algorithms compute the similarity of items or users according to a user-item rating matrix. However, traditional collaborative filtering algorithms face very severe data sparsity, which causes a discount of the performa
- PDF / 1,793,359 Bytes
- 8 Pages / 439.37 x 666.142 pts Page_size
- 101 Downloads / 240 Views
Abstract Traditional collaborative filtering algorithms compute the similarity of items or users according to a user-item rating matrix. However, traditional collaborative filtering algorithms face very severe data sparsity, which causes a discount of the performance of recommendation. In this paper, we proposed an improved clustering based collaborative filtering algorithm for dealing with data sparsity. We first clustered the users set into k clusters using K-means algorithm. Then we presented a formula to estimate those absent ratings in the user-item rating matrix and acquired a high density matrix. After that, we use the new rating matrix to calculate the similarity of items and predict the ratings of a target user on items which have not been rated and recommend Top-N items to the target user. We also implemented experiments and demonstrated that our proposed algorithm has better accuracy than traditional collaborative filtering algorithms. Keywords Collaborative filtering algorithm Data sparsity
Recommendation systems
K-means
Q. Wang Y. Liu (&) School of Communication and Information Engineering, Beijing Jiaotong University, Beijing 100044, China e-mail: [email protected] Q. Wang e-mail: [email protected] Q. Wang Y. Liu Key Laboratory of Communication and Information Systems, Beijing Municipal Commission of Education, Beijing Jiaotong University, Beijing 100044, China W. Cao China Information Technology Security Evaluation Center, Beijing, China e-mail: [email protected]
Y.-M. Huang et al. (eds.), Advanced Technologies, Embedded and Multimedia for Human-centric Computing, Lecture Notes in Electrical Engineering 260, DOI: 10.1007/978-94-007-7262-5_77, Springer Science+Business Media Dordrecht 2014
673
674
Q. Wang et al.
Introduction Recommendation systems can automatically recommend to users what they might be interested in. Usually we divide recommendation system algorithms into content-based algorithms [1, 2] and collaborative filtering algorithms [3–6]. Contentbased algorithms recommend to users items which are similar to what users have already bought or rated by analyzing the features of users or items. These algorithms can solve the problem called ‘‘cold start’’ and also won’t face the challenge of data sparsity because they don’t depend on the rating matrix. But they have a serious drawback that they can’t deal with pictures, video, music and other products difficult to be analyzed and extracted features from. On the contrary, collaborative filtering algorithms utilize a user-item rating matrix to calculate the similarity between users or items and then predict those items which have not been rated or bought depending on the ratings of neighbors which have high similarity with the target users. However, the number of items which each user has bought is usually less than 1 % of the total number of items in a site, which causes severe data sparsity and a decrease of the performance. In this paper, we proposed an improved clustering based collaborative filtering algorithm for d
Data Loading...
