Deconstructing Taste: Toward a Human-Centered AI Framework for Modeling Consumer Aesthetic Perceptions

TL;DR

Understanding and predicting consumer aesthetic taste is challenging due to its subjectivity and contextual nature. The authors propose a human-centered computational framework that links subjective style judgments with both designer-informed design patterns and machine-extracted visual features, demonstrated through a car-wheel aesthetics study. They show that both high-level design cues and low-level visual cues contribute to style perceptions, and that semantic alignment between consumer language and model-generated descriptions varies by style, supporting interpretable design guidance early in development. The framework enables designers to translate taste signals into actionable parameters, with potential applicability across domains beyond automotive aesthetics.

Abstract

Understanding and modeling consumers' stylistic taste such as "sporty" is crucial for creating designs that truly connect with target audiences. However, capturing taste during the design process remains challenging because taste is abstract and subjective, and preference data alone provides limited guidance for concrete design decisions. This paper proposes an integrated human-centered computational framework that links subjective evaluations (e.g., perceived luxury of car wheels) with domain-specific features (e.g., spoke configuration) and computer vision-based measures (e.g., texture). By jointly modeling human-derived (consumer and designer) and machine-extracted features, our framework advances aesthetic assessment by explicitly linking model outcomes to interpretable design features. In particular, it demonstrates how perceptual features, domain-specific design patterns, and consumers' own interpretations of style contribute to aesthetic evaluations. This framework will enable product teams to better understand, communicate, and critique aesthetic decisions, supporting improved anticipation of consumer taste and more informed exploration of design alternatives at design time.

Figures (13)

• Figure 1: Model of aesthetic experiences (taken from leder2004model). The multi-stage model shows how perceptual analyses, explicit classification, and cognitive mastering interact under the influence of prior experience, domain expertise, and personal taste. Early stages emphasize automatic, perceptual and affective processing, while later stages involve deliberate interpretation and evaluation. The proposed framework operationalizes the three components highlighted in red.
• Figure 2: Data collection pipeline involving human subjects. Starting from 1000 curated wheels, we derived high-level visual features from crowd-based feature annotations informed by designer-informed classifications. We used a subset of the wheel dataset to run multiple pairwise comparisons with human raters, which enabled subsequent analyses after converting pairwise ratings into Bradley-Terry scores.
• Figure 3: Pairwise comparison task. Participants selected one of two wheels, with each viewing a unique set of 240 wheel pairs randomly sampled from the 3,160 possible pairwise combinations.
• Figure 4: Wheel design image captions generated by OpenAI GPT-5.2. The top and bottom captions describe left and right wheel images in Figure \ref{['fig:pairwise_comp']}.
• Figure 5: Visualization of dominant line orientations in a wheel image. The histograms show the frequency distribution of line segments counted in each angle bin.