Examining Customer Satisfaction Through Transformer-Based Sentiment Analysis for Improving Bilingual E Commerce Experiences

This project explores customer satisfaction analysis in a bilingual e-commerce setting using transformer-based sentiment analysis models. The dataset consists of customer reviews originally in Chinese, converted to English for processing. The study employs three advanced transformer models—BERT-CNN, RoBERta-BiLSTM, and DistilBERT—to classify customer feedback into sentiment categories. The system includes modules for user registration, login, review classification, and logout, developed using a Flask backend and an HTML, CSS, and JavaScript frontend. This approach aims to improve understanding of customer opinions across language barriers by accurately detecting sentiments in translated text. The analysis helps identify patterns in satisfaction levels and provides insights into enhancing e-commerce experiences for users who interact in multiple languages. The project emphasizes simplicity in design while ensuring reliable sentiment classification through the use of modern natural language processing techniques. Results demonstrate the comparative effectiveness of the selected transformer models on the converted bilingual data. This work contributes to bridging language gaps in e-commerce feedback analysis, enabling better interpretation of customer sentiments and supporting improvements in service quality.

Keywords: customer satisfaction, sentiment analysis, transformer models, BERT-CNN, RoBERta-BiLSTM, DistilBERT, bilingual data, e-commerce, Flask, natural language processing.