A new uncertainty-aware similarity for user-based collaborative filtering

User-based Collaborative Filtering (UBCF) is a common approach in Recommender Systems (RS).Essentially, UBCF predicts unprovided entries for the target user by selecting similar neighbors.The effectiveness of UBCF greatly depends on the selected similarity measure and the subsequent choice of neighbors.This paper presents a new Uncertainty-Aware Similarity measure "UASim" which enhances CF by accurately calculating how similar, dissimilar, and uncertain users' preferences are.Uncertainty is a key factor of "UASim" that is managed in the neighborhood selection step of CF.Extensive experimental evaluation, conducted on Flixter, Movielens-100K, and Movielens-1M datasets, indicates that "UASim" shows better performance compared to many representative predefined similarity measures.The proposed measure demonstrates enhancements across various performance indicators, namely: Mean Absolute Error (MAE), Root Mean Square Error (RMSE), coverage, and the F-score.