Predictive model for customer satisfaction analytics in E-commerce sector using machine learning and deep learning

Abstract

In Vietnam's rapidly expanding e-commerce landscape, there is a critical need for advanced tools that can effectively analyze customer feedback to boost satisfaction and loyalty. This paper introduces a two-step predictive framework merging deep learning and traditional machine learning to analyze Vietnamese e-commerce reviews. Utilizing a dataset of 10,021 reviews on Tiki, Shopee, Sendo, and Hasaki between 2015 and 2023, the framework first employs fine-tuned deep learning models like BERT and Bi-GRU to extract aspect-based sentiments from reviews, tailored for the nuances of the Vietnamese language. Subsequently, machine learning algorithms like XGBoost predict customer satisfaction by integrating sentiment analysis with e-commerce data such as product prices. Results show BERT and Bi-GRU yield over 70% sentiment accuracy, while XGBoost achieves 80%+ satisfaction prediction accuracy. This framework offers a potent solution for discerning customer sentiments and enhancing satisfaction in Vietnam's dynamic e-commerce landscape.

Keywords

• Customer satisfaction
• E-commerce
• Machine learning
• Deep learning
• Predictive analytics
• Sentiment analysis