Insights from Amazon Reviews through Machine Learning-Aided Sentiment Analysis and NMF-based Topic Modeling

This study evaluates six machine learning models in sentiment analysis on e-commerce consumer reviews from three leading cosmetic brands. Among these, SVM, Naive Bayes and Random Forest are the most widely used conventional models in recent years, despite exhibiting certain limitations when processing sparse and non-linear text data. While gradient boosting models (XGBoost, CatBoost, and LightGBM) have demonstrated superior performance across various domains, their effectiveness in e-commerce sentiment analysis remains under-explored. To address this gap, this study compares these models to identify the most effective approach for sentiment analysis on e-commerce consumer reviews. The results indicate that SVM achieved the highest accuracy and F1 score, while CatBoost and LightGBM also exhibit robust performance under computational constraints. In addition, NMF-based topic modeling is applied to extract key product attributes that are frequently mentioned across different sentiment polarities. This combined approach not only enhances our understanding of consumer emotional responses but also provides actionable insights for product innovation and marketing strategies in the competitive beauty and personal care market.