Online shopping platforms generate millions of customer reviews every day. These reviews contain valuable opinions about products and services, but manually reading and analysing them at scale is not feasible. This is where Natural Language Processing (NLP) and sentiment analysis become important — they allow computers to automatically determine whether a review expresses a positive, negative, or neutral opinion. While progress in this area has been significant, several challenges remain. Traditional machine learning methods use statistical techniques like TF-IDF to represent text, which capture word patterns but miss deeper meaning and context. Advanced deep learning models like BERT understand language much more richly, but require large amounts of data and significant computing power. Most existing studies also work with a single dataset from one product category, limiting how well their models perform across different domains. Additionally, using star ratings to label reviews as positive or negative can introduce bias, and neutral reviews remain particularly difficult to classify accurately. This study proposes a combined framework that brings together traditional machine learning models and a fine-tuned BERT model to tackle these challenges together. Three different e-commerce datasets — customer feedback, clothing reviews, and Amazon product reviews — were merged to build a more diverse and representative training corpus. The system was also designed to perform Aspect-Based Sentiment Analysis (ABSA), which identifies opinions about specific product features such as price, quality, delivery, and packaging. A bias evaluation component was included to measure how much the star-rating labelling strategy affects model reliability. Results show that traditional machine learning classifiers such as Logistic Regression and SVM achieve approximately 92–93% accuracy, while BERT reaches around 90–91% under the current computational setup. The proposed framework is scalable, interpretable, and suitable for practical deployment in real-world e-commerce sentiment analytics.
Hussain et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: