Sentiment Analysis of Beauty Product Reviews in Bangladesh: A Comprehensive Study Using Transformer-Based Models

Sentiment analysis is crucial in analyzing consumers’ opinions when deciding future product directions. Beauty product companies can tailor their products to better meet customer needs by analyzing the detailed feedback in product reviews. Despite its importance, there is limited research focused on the sentiment analysis of beauty product reviews in Bangladesh. We have created a dataset of beauty product reviews to address this gap. Our dataset comprises approximately 11555 reviews collected from various platforms. We utilized seven models for sentiment analysis, including simple neural networks (SNN), 1D convolutional neural networks (1D CNN), long short-term memory (LSTM), bidirectional LSTM, a hybrid CNN + LSTM model, BanglaBERT and XLM-RoBERTa. Our analysis revealed that BanglaBERT and CNN produced the best results. Specifically, BanglaBERT achieved the highest Test Accuracy of 85.9% and demonstrated strong performance across all metrics with an F1-score of 0.8537 and a Precision of 0.8499, respectively. 1D CNN also exhibited excellent performance with a Test Accuracy of 85.83%, an F1-score of 0.8499 and a precision of 0.8446.