Agrupamento de grafos e sistema de recomendação: um estudo de caso das avaliações da Amazon

Abstract

In the pandemic, due to social distance and trade restrictions, online sales grew. Without the sellers, the site itself needs to make recommendations to customers só that they buy other products. Sales and valuation data and their relationships can be described as graphs, more specifically, bipartite networks and these can be used to recommend products to customers. The objective of this work is to identify groups of customers with similar preferences using the BRIM algorithm of graph clustering and, based on their preferences, recommend other products that may be of interest. Four scenarios were studied, considering the use of clustering and number of connections, and different steps of database cleaning. The best scenario indicates that users who rated only one product and those who rated all products with the maximum grade and duplicate ratings should be disregarded. The clustering proved to be effective with a modularity of 0.843. Because it requires a small quantity of data and is easy to implement, the methodology can contribute to increasing sales and avoiding the closing of small businesses. The limitations found were the time to generate the clusters and that the recommendation is only made for users who have performed more than one assessment. For this reason, many products are not even considered in the evaluation base due to cleaning and, therefore, the system ends up not recommending products that are rarely purchased or recently added.