Um estudo comparativo da acurácia de algoritmos de recomendação em sistemas de compras coletivas.

Abstract

Daily Deals Sites (DDSs) correspond to a specific website type designed to advertise offers (products, services or travel) at a significantly reduced prices, for a certain time, so the seller of the offers can make a large number of negotiations in a short period of time. To achieve this goal, the DDSs use marketing strategies ranging from advertising the offers on social network until to sending daily e-mails to registered users. However, the disclosure of the offers for most DDSs is not performed in a personalized manner, so all users receive the same set of daily offers. Thus, because of the lack of customization of this disclosure, users end up receiving a large amount of irrelevant or uninteresting offerings. Accordingly, we propose the study of a Recommender System that takes into account important aspects of users of DDSs. This aspects are defined by analyzing the real database of a company engaged in the group of buying domain. We evaluated four algorithms applied to data in this domain, two of these considered state of the art on recommendation, and discuss the results obtained from the experiments, also indicating which of those algorithms presents better efficacy, according to the metrics defined in this work. The evaluation of this work was performed by experimental means in partnership with company QueroDois, based in Ribeirão Preto - São Paulo.