Authors - Matheus Alcantara de Santana, Claudio de Souza Baptista, Andre Luiz Firmino Alves, Anderson Almeida Firmino, Gerson da Silva Januario, Roney Wellington da Silva Caldera Abstract - Driven by technological advances that make the purchase process more comfortable and efficient, e-commerce has grown every year, thus increasing the number of internet sales. Because of the large volume of products available from different suppliers, it becomes difficult to identify and group the same products to find the best deals. One of the most significant barriers is the existence of various descriptions of the same products. In this work, we use product matching with a heuristic created to help identify similar products to solve this problem. We employ several machine learning models, including BERT, and use a database of invoices from Brazil to carry out the experiments. At the end of the implementation, the model could classify the invalid products. The results of the classification task showed that our proposed solution is promising, with an f1 score of 92%.