Algoritmos de recomendação: propostas para avaliação das recomendações de itens únicos ou lista de itens

Recommendation algorithms play a role in improving and automating the task of making good recommendations. However, quantitatively measuring the quality of recommendations can be too complex. In this work, we propose two metrics to evaluate the result of recommendation algorithms: the Single Recommended Item Metric (SRIM) and List of Recommended Items Metric (LRIM). In the first, the purpose is to evaluate recommendation algorithms that return only one item, while the second has the objective of evaluating recommendations of multiple items. In this way, we apply the proposed metrics to evaluate such recommendation algorithms using a classic movie database. The results show that the AR2 recommendation algorithm provides more relevant recommendations than the AR2 algorithm according to the SRIM and LRIM metrics. In addition, we detected that the AR2 algorithm outperforms by just over 8% compared to AR1.