Analyzing Customer Sentiments In E-Commerce Reviews Using Machine Learning Models

Abstract

In the digital commerce landscape, online reviews are used to show the impact of purchasing decisions of the consumer. Positive reviews act as a social proof for encouraging new customers and increase the confidence for them to purchase an item. This study investigates sentiment analysis on Amazon product reviews using machine learning approaches. The dataset comprises 20,000 reviews obtained from Kaggle, encompassing positive, neutral, and negative sentiments. This imbalance dataset contains diverse sample of customer reviews. The class imbalance issues are sorted out to improve the prediction. The data was preprocessed using standard text cleaning steps (punctuation removal, stop word removal, lemmatization), followed by TF-IDF feature extraction with tuned parameters. The five machine learning models like Naive Bayes, Logistic Regression, Support Vector Machine (SVM), Random Forest, and K-Nearest Neighbors (KNN) were evaluated. Results show that SVM achieved the highest accuracy (91.4%) and F1-score (90.2%), outperforming other models. Evaluation was extended using per-class precision, recall, F1, macro-F1, and weighted-F1 to capture class-level differences, especially for the neutral class. Statistical tests validated SVM’s superiority over alternatives. These findings provide a foundation for real-time sentiment analysis in business intelligence.