<![CDATA[Indonesia’s Muslim women’s fashion market has expanded rapidly alongside e-commerce growth, generating massive volumes of online product reviews (OPRs) that remain underutilized for systematic product development. This study addresses a gap in the literature: while sentiment analysis can classify review polarity, term-level classification alone cannot translate consumer feedback into actionable design attributes for fashion products, a domain where tacit knowledge, material properties, and aesthetic judgment are central. A two-layer hybrid approach is proposed that combines computational sentiment extraction with expert semantic translation. In the first layer, 2,050 OPRs from three Indonesian Muslim fashion brands on Shopee were preprocessed and classified using a maximum entropy (MaxEnt) model, achieving 84.11% accuracy, 90.09% precision, and an F1 score of 89.95% on test data. In the second layer, ten experienced designers interpreted the MaxEnt output through structured interviews, translating raw sentiment features into design-relevant categories. Positive sentiment features clustered around product quality, material comfort, and design authenticity, while negative features concentrated on product-image discrepancies, poor fabric quality, sizing mismatches, and color inaccuracy. Designer interpretation uncovered semantic dimensions invisible to the classifier, yielding eight major product performance categories. This study contributes methodologically by demonstrating the necessity of a human-in-the-loop expert validation layer for sentiment-based consumer insight extraction in design-intensive domains, and practically by providing a framework for converting OPR data into product development inputs.]]>
