Predicting and explaining customer satisfaction: A deep learning and sentiment analysis of emotional impacts

This study examines customer satisfaction in retail shopping organizations by analyzing sentiment and emotional expressions in Yelp reviews. Utilizing a large dataset of over 500,000 reviews, the research applies keyword-based feature engineering to identify satisfaction from text, supported by a novel set of satisfaction-related keywords generated via ChatGPT. Deep learning models-Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN)-were trained on raw review texts to validate the keyword-based classification approach. Additionally, the NRC emotional lexicon was employed to extract emotional dimensions such as trust, fear, surprise, anticipation, joy, sadness, and disgust. Ordinary Least Squares (OLS) regression analysis revealed that sentiment, trust, and joy positively influence satisfaction, while fear, surprise, anticipation, sadness, and disgust negatively impact it. The consistency of results across two independently constructed datasets underscores the robustness of the methodology. This integrative approach combining deep learning, lexicon-based emotion analysis, and regression modeling provides nuanced insights into the emotional drivers of customer satisfaction in retail contexts, providing a diagnostic framework and foundational insights into emotional drivers, which are essential for improving consumer experience and loyalty.