Integrating Information Retrieval with BERT: Advanced Sentiment Analysis Framework for E-Commerce Product Reviews
M. Kathiravan
Abstract
This research study presents a novel sentiment classification system specifically designed for analysing e-commerce reviews. The methodology is based on the Distil-BERT model. It effectively handles linguistic intricacies in several languages, including idioms and sarcasm, unlike conventional rule-based systems. The model demonstrates proficiency in twelve languages, such as English, Dutch, German, French, Spanish, and Italian, by employing the Amazon Review Dataset. This system is integrated with online scraping tools and hosted on Hugging-Face. It efficiently accumulates and analyses consumer feedback from Amazon, achieving an impressive precision rate of 90%. The multilingual model in question plays a vital role in sentiment analysis, providing deep insights for the e-commerce business and representing a significant advancement in market research methods.