Item-based collaborative filtering with fuzzy vector cosine and item directional similarity

Collaborative filtering algorithm is applied successfully in the field of e-commerce recommendation and personalization. But it faces the sparsity and scalability problems which deteriorate recommendation performance greatly. A combined algorithm employing fuzzy vector cosine similarity and Pearson correlation with item directional similarity is proposed. The rating matrix is converted to a fuzzy matrix which provides a new similarity measure to relax the constraints in similarity calculation. Moreover, the similarity scale is adjusted by item directional similarity to weaken the dissimilar neighbors´ noise. Finally, the experimental result of the proposed algorithm based on MovieLens data set is given. And the result shows the proposed algorithm has good prediction accuracy and is robust to the neighborhood size.