On explainable flexible fuzzy recommender and its performance evaluation using the akaike information criterion

Abstract

In the paper, fuzzy recommender systems are proposed based on the novel method for nominal attribute coding. Several flexibility parameters - subjects to learning - are incorporated to their construction, allowing systems to better represent patterns encoded in data. The learning process does not affect the initial interpretable form of fuzzy recommenders rules. Using the Akaike Information Criterion allows evaluating the trade-off between a number of rules and interpretability which is crucial to provide proper explanations for users.