Paper Contents
Abstract
The rapid growth of online shopping has led to an explosion of user-generated content, particularly in the form of product reviews. These reviews provide valuable insights into customer satisfaction, product quality, and user expectations. However, the sheer volume of such feedback makes manual analysis nearly impossible. As a result, automated sentiment analysis has become an essential tool for interpreting opinions embedded in textual data. Sentiment analysis, a key area within Natural Language Processing (NLP), focuses on identifying the emotional tone behind written content and categorizing it as positive, negative, or neutral.Earlier approaches to sentiment classification relied on traditional machine-learning models such as Logistic Regression, Nave Bayes, and Support Vector Machines. These models depend heavily on handcrafted features like Bag-of-Words and TF-IDF representations. Although they work reasonably well for simple datasets, they struggle to understand context, sarcasm, long text sequences, and complex linguistic patterns. Deep-learning methods like CNNs and LSTMs improved performance by learning richer feature representations, but their ability to capture bidirectional context and long-range dependencies remains limited.The introduction of Transformer-based architectures reshaped modern NLP by enabling models to learn contextual relationships using self-attention mechanisms. Among these, BERT (Bidirectional Encoder Representations from Transformers) has become a leading model due to its capability to process text in both forward and backward directions simultaneously. Its contextual understanding allows it to recognize subtle sentiment cues, implicit opinions, and mixed emotions far more effectively than earlier models.This research aims to develop a sentiment analysis system using BERT to classify Amazon product reviews and evaluate its performance against classical machine-learning and deep-learning baselines. The study involves cleaning and preprocessing review data, fine-tuning BERT on domain-specific text, and assessing the model using metrics such as accuracy, precision, recall, and F1-score. In addition, the research integrates interpretability tools like SHAP and LIME to understand the factors influencing model decisions. Through this work, the paper demonstrates how advanced NLP models can offer reliable, scalable, and context-aware solutions for analyzing large volumes of customer feedback.
Copyright
Copyright © 2025 Pragati Malhotra. This is an open access article distributed under the Creative Commons Attribution License.
