An Entropy-Based Composite Indicator for Evaluating the Effectiveness of Recommender System Algorithms

Abstract

The problem of forming a composite indicator for evaluating the effectiveness of recommender system algorithms is considered. A novel composite indicator is proposed by combining individual metrics using the entropy method. The testing base of this study consists of 12 algorithms (on the one hand) and 3 datasets (on the other). For each algorithm–dataset combination, we calculate partial criteria used in evaluating recommender systems. According to the results presented below, the composite indicator is an effective tool for evaluating the performance of recommender system algorithms. As is shown, the performance of the algorithms varies depending on the size and other basic characteristics of a particular dataset. This indicator can be used to develop more efficient algorithms and their ensembles as well as to optimize hyperparameters and improve the quality of recommendations.