Collaborative filtering recommendation algorithm based on sample reduction

Personalized Recommender System has become an important research field in Electronic Commerce, and the clustering of customers is the basis to produce the recommendation. The customers clustering in recommender system has some unique characters, such as the extreme sparsity of user rating data and huge sample space. Traditional Collaborative Filtering (CF) algorithm works poorly in this situation. To improve the quantity of recommending, sample reduction method is proposed in CF schema to lessen the sample space in both row and column aspects before carrying on the classification. It has the following features: (1) a Restraint Function is introduced into the basic CF model to solve the problem of sparsity of user rating data; (2) selective sampling and look-ahead framework are combined, based on the nearest neighbors algorithm, to reduce the number of samples while maintaining the quality of classification. At last, experiments are designed on the basis of MoveLens data set; recall and precision are applied as evaluating guidelines. Compared with general CF, the proposed algorithm has higher quality of recommendation.