Classificação automática de marcas utilizando algoritmos de aprendizado de máquina

Abstract

This work presents the development of an automatic brand classifier using machine learning algorithms, with the objective of identifying product brands based on textual descriptions. Automatic product categorization is a significant challenge in the retail and ERP sectors due to the large volume of unstructured textual data. By utilizing Natural Language Processing (NLP) techniques and textual representations such as Bag of Words (BoW), TF-IDF, Word2Vec, and FastText, the study applied models like Logistic Regression, Random Forest, Naive Bayes, and SVM to classify the brands. The dataset used includes 517,262 product descriptions, categorized into 1,964 distinct brands. Experimental results indicate that the combinations of BoW and TF-IDF with Logistic Regression and SVM models achieved the best performance, reaching an accuracy of 94%. On the other hand, techniques like Word2Vec showed inferior performance, suggesting that count-based vectorizers remain more efficient for this type of task. The study concludes that automatic brand classification can be significantly improved with the application of supervised models and recommends the use of more sophisticated representations for future enhancements.