Recomendação de produtos através do método de Market Basket Analysis aplicado ao cenário de Big Data

Abstract

Market Basket Analysis (MBA), or Análise de Cesta de Mercado in Portuguese, involves extracting a large number of product associations from transactional databases. Originally used to analyze shopping baskets in markets, it is not limited to this scenario and can be applied to any business that sells products. Recommendation systems, widely employed in prominent e-commerce platforms like Amazon and Mercado Livre, are commonly grounded in Data Mining and MBA techniques. This study presents a proposal for an adaptable recommendation system, developed to operate in both digital and physical environments of an extensive retail network. This work proposes an adaptable recommendation system for both digital platforms and physical stores of a large retail network using the MBA technique. The differentiating factor of this approach lies in its ability to make product recommendations on a large scale, optimizing the processing of large volumes of data through tools and distributed processing strategies, which were essential for resolving challenges related to big data. The motivation for this development stems from the observation that most existing research focuses on recommendation systems suitable only for small data quantities, in addition to aiding other researchers in addressing challenges in this scenario. It was possible to understand the relationships between recommended products based on region, sectors, and seasons, where behaviors between the northern and southern regions of the country were observed, products from the same sector strongly related, weather events influencing recommendations, and the recommender understanding product variations and recommending them accordingly. The effectiveness of the proposed system was assessed with an accuracy estimate of 22.53% for the digital channel and 31.28% for physical stores considering december/2023.