Research on personalized customization of interior design driven by artificial intelligence

Abstract

The traditional interior design industry is constrained by high labor costs, long design cycles, and difficulty in accurately capturing diverse and highly personalized user requirements. This study presents an AI-driven personalized interior design system that integrates multimodal attention, deep reinforcement learning, a conditional generative adversarial network (cGAN), and explainable AI. The design process is formulated as a sequential decision problem, enabling the system to generate and refine three-dimensional layouts, furniture configurations, and soft-furnishing schemes while satisfying physical constraints and aesthetic rules. Experiments on four public datasets and a user study involving 300 participants show that the proposed system achieves an IoU of 0.891, a style-consistency score of 0.862, and a functional-satisfaction score of 0.884. Among the 150 participants assigned to the proposed system, the mean overall satisfaction score was 4.25/5, and 88.7% indicated that they would recommend the system. These results demonstrate the feasibility of an efficient, end-to-end, and interpretable framework for personalized interior design.