Título: Sentiment Classification of Customer Reviews Using LSTM and BERT -based Transformers
Autores: Eduardo Furtado de Simas Filho, Carlos Eduardo da Silva Cerqueira, Felipe Guimarães Izidoro Freire & Ivon Luiz Correia Martinez Garcia
Resumo: In today’s digital landscape, online customer reviews play a crucial role in shaping consumer decisions and business strategies. However, the vast volume of reviews makes manual sentiment analysis impractical. This paper presents a comparative study of Natural Language Processing (NLP) techniques for sentiment classification in Portuguese-language e-commerce reviews. We evaluate multiple approaches, including Deep Neural Networks (DNNs), Long Short-Term Memory (LSTM) networks, and fine-tuned BERT-based transformers, using a Brazilian e-commerce dataset. Despite challenges such as inconsistencies between review text and assigned ratings, our findings show that effective preprocessing paired with LSTM architectures yields competitive accuracy (88.24%), while fine-tuning a Portuguese BERT model further improves performance (91.07% accuracy). The proposed methods provide scalable solutions for businesses to analyze customer sentiment and enhance product offerings.
Palavras-chave: Sentiment Analysis; Natural Language Processing; Customer Reviews; Deep Learning; BERT.
Páginas: 7
Código DOI: 10.21528/CBIC2025-1188426
Artigo em PDF: CBIC_2025_paper1188426.pdf
Arquivo BibTeX:
CBIC_2025_1188426.bib