Consumer Behavior

Revisiting Fashion Product Reviews: AI-Driven Analysis of the Pre-Purchase vs. Post-Purchase Divide

---

Abstract

Consumer-generated online reviews are invaluable for text mining research and industry insights. However, existing studies often overlook a key distinction: the divide between pre-purchase (e.g., expectations) and post-purchase (e.g., experiences) content within a single review. Drawing on expectation-confirmation theory, we propose this is critical for fashion, where pre-purchase aspirations drive purchase intention while post-purchase experiences determine satisfaction. Using real-world reviews from online retailer Musinsa, we trained a language model under three experimental conditions of training data to test whether segmenting reviews into pre-purchase and post-purchase content improves sentiment prediction (positive or negative). Dictionary-based natural language processing (NLP) was applied to segment reviews. Finally, we tested our approach on 3-star reviews labeled by human experts. Evaluation results show that our approach improved methods that do not make this distinction. Findings contribute to fashion NLP by advancing nuanced understanding of consumer-written narratives, and rethinking methods in the context of fashion consumers.

Keywords: sentiment analysis, Online reviews, text mining, natural language processing

How to Cite:

Rhee, H. & Zhao, L., (2025) “Revisiting Fashion Product Reviews: AI-Driven Analysis of the Pre-Purchase vs. Post-Purchase Divide”, International Textile and Apparel Association Annual Conference Proceedings 82(1). doi: https://doi.org/10.31274/itaa.21981