Toward an AI-Learnable Product Form Evaluation: Linking Design Language and Quantifiable Parameters

Product form shapes how products are recognized, interpreted and chosen. However, research on form is fragmented across levels of analysis: product meaning and brand recognition, perceptual descriptors and semantic differentials, and geometric or computational measures of form. The paper argues that this fragmentation makes it hard to build datasets or AI-learning pipelines that designers can understand. A model may learn from pixels or geometry alone, but without a structured representation of category, style and form indicators, what it learns is difficult to interpret and validate in a design context. The literature review finds no cohesive framework that connects product type and style to design-relevant indicators, measurable parameters and normalized scores. The research question is how qualitative design knowledge about product form can be structured into a systematic evaluation framework that is interpretable to designers and usable for future AI learning.

A conceptual study in three steps:

• Literature synthesis across product semantics, form perception and aesthetic judgment, shape and silhouette recognition, symmetry and curvature preference, unity-in-variety and processing fluency (the idea that stimuli that are easy to process are liked), visual complexity and computational aesthetics, geometric continuity and "fairness" (how smoothly curvature changes on CAD surfaces), and industrial design pedagogy on massing and hierarchy.
• A four-level hierarchical framework (Figure 1): (1) product type, the contextual baseline; (2) design style, one or more labels such as minimalistic, biomorphic, geometric, mechanical or iconic, treated as structured descriptors rather than fixed categories; (3) form indicators, such as silhouette recognizability, symmetry and axiality, curvature character, unity in variety, visual complexity, surface or edge continuity, massing hierarchy, balance, proportion and part segmentation, each grounded in the reviewed literature; and (4) measurable parameters, converted into normalized scores either computationally (e.g., symmetry distance, edge density, identification rate) or through structured expert judgment with Likert-scale rubrics for indicators that resist automated measurement (e.g., massing hierarchy clarity, unity-in-variety).
• An illustrative chair example (Figure 2) and an indicator-to-metric mapping table (Table 1).

• Proposed framework: a four-level structure linking product type, design style, form indicators and measurable parameters with normalized scores, which the paper describes as a conceptual integration model rather than an entirely new theoretical construct.
• Proposed principle (traceability): every score should stay connected to the parameter it came from, every parameter to a named indicator, and every indicator should stay interpretable within its product-type and design-style context.
• Illustrative example: a chair, tentatively labeled minimalistic, is evaluated on five indicators (silhouette recognizability, symmetry and axiality, proportion system, curvature character, visual complexity). Each is linked to a candidate metric family rather than one fixed measure, e.g., identification rate or shape distance relative to an archetypal chair contour, and symmetry distance and dominant-axis count.
• Proposed indicator-to-metric mapping (Table 1): for example, curvature character maps to curvature histogram, corner density and radius distribution; visual complexity maps to part count, edge density and contour change frequency; proportion system maps to height-width-depth ratio and part-to-whole ratio.
• Conceptual contribution (as argued by the authors): the framework bridges qualitative design language and quantitative representation, supports more consistent evaluation (design review, design education, dataset annotation) even before automation, and supports an interpretable approach to AI-oriented design research.

The paper states a threefold contribution: a synthesis of literature on product semantics, appearance, style, form perception and quantification; a hierarchical framework linking category and style labels to named form indicators and candidate metrics; and a demonstration through a chair example. The authors state that the study is primarily theoretical and conceptual, and that it reorganizes existing design theories rather than introducing a new construct. The framework has not yet been validated through expert studies or user-based evaluations. The indicator set will likely need refinement across product categories, some indicators may require expert judgment or hybrid annotation rather than automatic extraction from images or CAD models, and the weighting of indicators remains unresolved. Color, material and finish (CMF) are excluded and left for future work. The paper outlines a staged agenda that reserves empirical validation and AI experimentation for later studies.