Author : Numan Ali, Sadia Ramzan, Tasawar Ali and Muhammad Usman

In recent years, customers have increasingly provided essential feedback, opinions, and recommendations for internet retailers. This article aims to develop an automated comment analyzer. We present an automated solution for analyzing and classifying customer comments derived from Amazon data domains, capable of managing a substantial volume of reviews. Supervised learning classifiers, specifically Naive Bayes (NB), Support Vector Machine (SVM), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM), are employed to categorize comments as positive or negative. This study utilizes three variations of Naive Bayes models, including Support Vector Machine, Multinomial Naive Bayes, and Complement Naive Bayes, for sentiment analysis of e-commerce reviews. The system is tested and evaluated using real-time data, including product reviews from Amazon’s website, specifically analyzing 10,000 customer reviews spanning various items. Data preprocessing techniques, such as lowercase processing, stop word removal, punctuation removal, and tokenization, enhance the usability of the collected data for analysis. The models were trained on this cleaned dataset to identify and classify customer sentiment as positive or negative. The machine learning algorithms CNB, MNB, BNB, and SVM achieved accuracies of 80.00%, 79.90%, 79.35%, and 81.25%, respectively, while the deep learning algorithms GRU and LSTM obtained accuracies of 80.6097% and 76.2619%, respectively. Although the SVM model demonstrated greater accuracy than the deep learning models, it exhibited significantly slower execution times. Our findings indicate that deep learning approaches yield superior results for categorizing consumer attitudes toward products.

Keywords: Sentiment Analysis; Machine Learning; Customer Feedback; E-commerce Reviews; Deep Learning Models