Paper Title: Advanced large language models for text analytics and consumer opinion analysis
Author: Nilesh Jain
Corresponding Author: Nilesh Jain (nileshjainmca@gmail.com)/India
Abstract
Large e-commerce sites receive many customer reviews, offering valuable insights into consumer experiences, preferences, and opinions. However, unstructured, high-volume review text makes manual analysis inefficient. This study introduces a Textual Analysis and Consumer Sentiment Classification Approach based on a BERT framework, using the Amazon customer review dataset of 30,841 reviews. The proposed structure handles cleaning, tokenization, feature preparation, class balancing, contextual representations, and BERT-based sentiment classification. The data is first stratified to ensure the two classes are evenly distributed, then split 80/20 into training and testing sets. Then, only the training data is balanced using SMOTE. The BERT model is trained to understand word context and categorize reviews by sentiment. The experimental results show the proposed model has an accuracy (ACC) of 95.94%, precision (PRE) of 95.91%, recall (REC) of 95.97%, and F1 score (F1) of 95.92%. The results show that BERT can model context-aware information in customer reviews and automatically classify sentiment and extract valuable insights from customer feedback in large-scale e-commerce scenarios.
