Explaining Recommender Systems by Evolutionary Interests Mix Modeling

Piotr Lipinski

Abstract: This paper focuses on explaining the results of Recommender Systems, that aim at suggesting, for a given user, the most accurate products, among a given set of available products, and modeling how different types of user activities, such as based on user interests in different categories of products, affect the results of the recommender system. It proposes an evolutionary approach to interests mix modeling that defines the relation between the characteristic of the user ratings and the composition of the list of the recommended products. Computational experiments, performed on some selected benchmarks derived from the MovieLens dataset, confirmed the accuracy and efficiency of the proposed approach.