Explainable Human–AI Co-Design for 3D Generation: A Knowledge-Driven Framework Leveraging User-Generated Content

In human–AI co-design scenarios, designers often struggle to interpret, trust, and effectively control opaque generative processes, limiting collaborative creativity. To address this challenge, this study proposes E3D, an explainable human–AI co-design framework that integrates knowledge-driven reasoning, generative 3D modeling, and visual explainability. The framework transforms UGC from social media into structured design knowledge through sentiment-aware semantic parsing and large language model–based prompt engineering. To enhance interaction transparency, E3D incorporates a Grad-CAM–based visual reasoning module and multi-dimensional feedback mechanisms, allowing users to inspect how preferences, constraints, and contextual cues influence generated forms. A controlled user study demonstrates that explainable feedback significantly improves design quality, modeling accuracy, and task efficiency, while increasing users’ perceived interpretability, trust, and sense of control. Ultimately, thefinding of this work is that combining multimodal inputs with actionable, explainable feedback fundamentally improves interaction outcomes in human–AI co-design.