Sentiment Analysis of Bangladeshi E-Commerce Site Reviews Using Machine Learning Approaches

In the context of Bangladesh, the E-commerce sector is experiencing continuous growth, particularly during the global crisis. Amidst the plethora of available platforms, Daraz has emerged as the most successful marketplace, offering users a wide array of shopping options. However, the abundance of reviews and comments on this online platform presents a challenge for consumers trying to make optimal choices. This research focuses on systematically categorizing positive and negative reviews to enhance user decision-making. To achieve this objective, a range of classifiers, including Multinomial Naive Bayes, Logistic Regression, Decision Tree Classifier, Random Forest Classifier, K Neighbors Classifier, and Support Vector Machine with different kernels, were employed. The dataset underwent thorough cleaning, followed by the application of Term Frequency-Inverse Document Frequency (TF-IDF) with Principal Component Analysis (PCA) to enhance feature representation. The findings of this study indicate that the Multinomial Naive Bayes classifier, especially when utilizing Bigram and Trigram features, outperformed other classifiers, demonstrating superior accuracy. The implementation of this classifier holds significant promise for assisting businesses operating on various platforms, enabling them to distinguish between positive and negative reviews effectively. Consequently, this approach empowers businesses to furnish customers with valuable insights into the quality of products, contributing to a more informed and confident consumer base.