关键词: 连衣裙推荐, 体型特征, 综合评价模型, 虚拟试衣, 交互效应, 服装个性化推荐

Abstract:

Objective To address the inaccuracy in clothing personalization systems resulting from the neglect of body type factors, this study proposes a comprehensive dress evaluation model that integrates anthropometric characteristics and design element interactions. The model aims to enhance recommendation accuracy through the quantification of body-type-specific preferences and the analysis of their interactions.

Method Five female body types (inverted-triangle, rectangular, triangular, hourglass, and oval) were identified using virtual fitting technology combined with anthropometric classification. Key dress design elements, namely silhouette, fit, waist position and dress length, were extracted through morphological analysis. Experimental samples combining body types with design variations were generated, and subjective evaluations were collected through two scenario-based experiments. Multivariate repeated-measures ANOVA was applied to analyze the influence of body type on preference patterns, with interaction effects quantified using effect size and correction values. A comprehensive evaluation model was constructed by integrating feature weights, preference scores, and interaction adjustments.

Results The study used a multifactorial repeated-measures ANOVA to assess consumer preferences for dress design attributes. The results revealed significant main effects for both silhouette and fit; silhouette exhibited the largest effect size, and all two-way interactions were significant despite their relatively small effect sizes. The average preference score across all combinations was 2.81. The X-tight combination received the highest score (3.68), while the Y-loose combination scored the lowest (2.14). The line chart revealed considerable variation in ratings across silhouettes compared to the relative stability observed across fit levels, indicating silhouette's substantial impact on consumer preference. For silhouettes, consumers consistently preferred X-line and A-line, while fitted styles were preferred in terms of fit. Body type also influenced consumer preferences, with rectangular types showing the greatest sensitivity to changes in silhouette, and inverted-triangle types being the most responsive to fit.

The multifactorial repeated-measures ANOVA revealed a significant main effect for waist position, a significant two-way interaction between dress length and body type, and a significant three-way interaction among waist position, dress length, and body type. The overall mean rating across all combinations was 3.19. Among all combinations, dresses with mid-waist and mid-length received the highest rating (3.30), whereas high-waist, short-length dresses received the lowest (3.03). Mid-waist designs were consistently preferred over high-waist designs across all body types. However, dress length preferences varied significantly: inverted-triangle types favored shorter lengths, while oval and triangle types preferred medium lengths. Although variations in waist position or dress length resulted in only slight differences within a single body type, the same combination of waist position and length received significantly different scores across different body types.

The interaction effect between silhouette and fit was significantly moderated by body type. A complex three-way interaction among waist position, dress length, and body type led to significant differences in composite scores. Different body types demonstrated significantly different prioritization of style elements, including silhouette, fit, waist position, and dress length. To quantify these differences, feature weights were used to represent the relative importance of specific style element categories for each body type. To measure the deviation between actual composite scores and theoretical scores caused by interaction effects, an interaction adjustment value was introduced. Therefore, a comprehensive evaluation model that considers individual body type characteristics is necessary to provide tailored clothing style recommendations for consumers.

Conclusion Body type significantly influences the interaction between silhouette and garment fit. A complex three-way interaction exists among waist position, dress length, and body type, resulting in notable differentiation in composite scores. These findings confirm the necessity of incorporating body-type-specific parameters in predictive models to account for attribute interactions. Model validation demonstrated 72.3% average alignment between predicted and observed preferences through weighted calculations and interaction adjustments. This research establishes quantitative relationships between anthropometric characteristics and design preference patterns, providing a methodological framework applicable to other apparel categories. Statistical evidence supports the inclusion of interaction modifiers and body-specific weights in fashion data analysis systems to enhance predictive accuracy.

Key words: dress recommendation, body type characteristic, comprehensive evaluation model, virtual fitting, interaction effect, clothing personal recommendation