Sentiment Analysis of Reviews for E-Commerce Applications

This paper presents the application of sentiment analysis to reviews of e-commerce platforms, aiming to uncover consumer opinions and enhance decision-making for both businesses and customers. With the rapid growth of online shopping, understanding customer sentiment has become essential for improving user experience and optimizing product offerings. By employing Natural Language Processing (NLP) techniques including CountVectorizer, TF-IDF, N-gram, and five different Machine Learning (ML) algorithms including Multinomial Naïve Bayes (MNB), Logistic Regression (LR), Stochastic Gradient Descent (SGD), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), we analyze a substantial dataset of consumer reviews from Amazon e-commerce application. Our findings explain the impact of two different approaches for data labeling on model performance. Models implemented with three sentiment classes-positive, neutral, and negative-perform better compared to those implemented with two classes-positive and negative. RF and XGBoost models demonstrate superior performance when using TF-IDF and CountVectorizer techniques across three sentiment classes. The RF model achieves 91 % accuracy with both TF-IDF and CountVectorizer, while the XGBoost model also achieves 91% accuracy with CountVectorizer and 90%with TF-IDF. Additionally, SGD with TF-IDF and CountVectorizer outperforms in two sentiment classes, achieving 80% accuracy with both techniques.