Fine-Tuning Large Language Models for Multi-Task Consumer Data Analysis in Fashion Design Process: A Case Study of Chinese Women's Fashion Market

Abstract

Artificial intelligence (AI) is rapidly growing within the fashion industry, with current attention primarily focused on image  transformation and generation. However, the application of text comprehension AI in design processes, particularly in market research, remains insufficiently explored. This research fine-tunes RoBERTa models to construct an analytical framework including data cleaning, sentiment analysis, and topic classification for Chinese women's fashion analysis.The research analyzed 30,796 user comments from Bilibili. The fine-tuned models achieved strong performance: 95% accuracy for data quality classification, 97.65%
for sentiment analysis, and F1-scores ranging from 0.70 to 0.97 across nine topic categories.
Analysis of 6,029 high-quality comments revealed that 89.1% of consumers expressed negative or neutral sentiments, with size fit (43.5%) and gender differences (41.3%) being main concerns. The research identified nine systematic industry challenges, including size standards deficiencies, design practices that enforce traditional gender norms at the expense of functionality, and unfair pricing practices.This research shows that fine-tuning Large Language Models works for fashion processes analysis, providing evidence for widespread consumer dissatisfaction. The research fills the gap in applying fine-tuned LLMs to fashion design processes while demonstrating new ways for integrating fashion education with AI, contributing to digital transformation in fashion education and industry development.