Visual Preference Modeling and Optimization in Graphic Design via Feature Encoding and Apriori Rule Mining

Abstract

Graphic design currently lacks computationally accurate methods for identifying and optimizing user visual preferences, hindering personalized and precise design outcomes. This paper presents a computational method integrating multidimensional feature encoding and Apriori rule mining to extract interpretable visual preference patterns from user feedback, enabling targeted design optimization. A multidimensional feature matrix encompassing color, layout, font, and graphic structure is constructed and encoded into a standardized preference dataset derived from 1280 user-selected design samples. The Apriori algorithm extracts high-confidence association rules linking visual element attributes to user preference outcomes, filtering representative combinations with minimum support 0.05 and confidence 0.6. Rule sets are vectorized to represent structured user group preferences, and a preference pattern map is generated via K-means clustering with five clusters, achieving an average silhouette coefficient of 0.63. High-confidence rules are embedded as constraints in an automated design generation module, reconstructing visual solutions aligned with user preferences, validated at 89.2% extraction accuracy under support threshold 0.09. Experimental validation confirms 89.2% preference extraction accuracy at support threshold 0.09 and silhouette coefficient 0.63 for five clusters, demonstrating effectiveness and adaptability in preference modeling and optimization.