AN INTEGRATED HYBRID DEEP LEARNING MODEL FOR RETAIL SENTIMENT ANALYSIS

Abstract

The boom in online shopping websites has given rise to the growing demand of accurate sentiment analysis for finding out actionable intelligence from customer reviews. In this paper, we introduce a hybrid sentiment analysis system consisting of the rule-based (VADER) and the deep learning- based transformer model (RoBERTa) to categorize sentiment into three categories, i.e., positive, neutral and negative. The approach employs VADER for processing short reviews in near real time and fine-tunes RoBERTa to model intricate linguistic patterns across longer ones. For model training and testing, we employed Flipkart customer review datasets. We get 1281 jokes giving a precision of 89.2 %, recall of 88.1 % and F1 - score performance of 88.6 %. Power BI is utilized in the visualization of the predictions from models through interactive dashboards which gives live marketing insiughts as well. The hybrid system can deliver a more accurate and scalable sentiment classification, which is invaluable for organisations in the context of the optimization of their customer engagement strategies.