Abstract: The ratings of items based on the similarities between items are predicted by traditional item-based collaborative filtering methods However, the selections of the similar ones are suffering from limited scalability and accuracy. A distributed collaborative filtering method was proposed. This method clusters items into several blocks using fuzzy blocking, and performs comparisons solely among the items within each block. Additional efficiency enhancements can be achieved through the pruning of the similar relationship graph:edges between items that are not likely to be similar can be removed from the graph. It divides this graph into multiple smaller partitions from each which similarity degrees between items is calculated efficiently in parallel. Experiments show that the proposed method can improve the recommendation scalability and accuracy.

Key words: recommender systems, personalized recommendation, collaborative filtering, data blocking, fuzzy clustering