Collaborative filtering (CF) is considered as one of the most popular and widely used approaches in recommendation systems. CF makes automatic recommendations based on the similarity between users (user-based) or items (item-based) in the system. In this respect, various machine learning techniques were used to create model-based CF methods. However, most of the previous works do not consider the imperfections in the users’ ratings. Thus, in this paper, we tackled the issue of creating a rule-based CF model dealing with evidential data, i.e., data where imperfection is represented and managed thanks to the belief function theory. We proposed a novel method named ECFAR that learns recommendation rules from a soft rating matrix and uses them to make predictions. To assess the reliability of our method, we conducted various experiments on a real-world data set. The experiments show that our proposed method produces satisfying results compared to existing solutions.
