Pattern Detection in e-Commerce Using Clustering Techniques to Explainable Products Recommendation

Recommender systems in e-commerce websites have allowed customizing the delivery of products and services according to each consumer’s preferences. Nowadays, there is an increased interest in that the recommendations provided by these websites are explainable. In this paper, we present the use of clustering techniques to find relevant aspects of products and identify relationships and hidden patterns in the data. These patterns are considered for separating the products into groups, which then are used to make explainable recommendations according to the products’ metadata and ratings are given by users to them. Experiments are carried out with three clustering algorithms: K-means, DBSCAN, and Fuzzy C-Means, their results are compared using the silhouette coefficient. K-means and Fuzzy C-Means algorithms presented better performance with the dataset used. The K-means algorithm finds three product segments: the most expensive and moderately sold products; the cheapest and least sold products; products with regular prices and best-selling. For the recommendation, the item-based collaborative filtering approach is used to represent products that users liked, using the information of product rating and the group of the product to be recommended.