Applying User-Favorite-Item-Based Similarity into Slope One Scheme for Collaborative Filtering

Collaborative Filtering (CF) is one of the most effective technologies in making rating prediction for recommender systems. Traditional user-based CF methods take user-item matrices as input and compute the prediction value based on user similarity. The computation of similarity between each user pair requires exact matching on item set by finding a minimum number of same items rated by both users. However in real-world scenario, the input matrix is very sparse, therefore the amount of required information is far from enough to compute reliable similarity. To deal with this so-called data sparsity problem, we introduce a novel user similarity measure based on user favorite items (UFI) and apply it into Slope One scheme for rating prediction. The experimental results suggest that UFI user similarity measure is effective in user-based approaches and UFI-based Slope One algorithm can provide better quality of prediction than best available similarity-based CF algorithms.