Abstract: Collaborative filtering has been widely used in E-Commerce recommendation systems,but the sparsity of data affects the quality of collaborative filtering recommendation.A two-stage collaborative filtering algorithm is proposed based on random walking and cluster-based smoothing.For off-line stage,calculate the correlation betweenitems,suggest anew method which describes the correlation between items by cumulating weightedtransition probability of each step.Cluster items according to the item correlation matrix,then smooth the unrated data by using clustering information.For on-linestage,search the target item's neighbors according to the correlation between items cumulated during the off-line and predict.This method can enhance the description of the correlation between items.The experiment results illustrate that searching neighbors according to the item correlation matrix will become more accurate,which can effectively relieve theimpact of sparse data and improve the quality of recommendation.

Key words: collaborative filtering, random walk, correlation description, cluster-based smoothing, MAE