A fuzzy hybrid approach to enhance diversity in top-N recommendations

In recent researches, recommender systems performance was restricted in their ability to predict unseen items by users and recommending them those with the highest predicted ratings. To make good predictions it is not sufficient to judge the performance, because a good recommender should offer useful and diverse items that fit to different interest choices and tastes of users. Unfortunately, the lack of diversity and the stability in recommender systems over the user's profile dynamicity become a major challenge. In this paper, we propose a fuzzy hybrid diversified recommendation system able to generate multi-taste recommendations depending on the user's profile variation. Due to the fuzziness and uncertainty in user's profile, the system allows users belonging to different clusters using a fuzzy-based collaborative filtering combined with a content-based filtering algorithm. To identify the user's neighbourhood, a novel similarity measure is proposed. To increase diversity in recommended lists, subsets from Top-N recommended lists in similar clusters, are selected according to users membership degrees. Several experiments are conducted on Movilens dataset to prove the proposal's effectiveness.