Exploring developmental pathways in LLM-empowered affective design

Large Language Models (LLMs) augmented emotion research is emerging as a pivotal hotspot in the fields of affective computing and human–computer interaction. However, existing studies focus predominantly on practical applications, lacking systematic reviews from a design perspective. This paper systematically synthesises 72 LLMs-assisted emotion studies and 83 traditional Affective Design (AD) publications, constructing a reusable Emo-LLMs corpus. We clarify how LLMs reshape the four core steps of AD, namely scenario and task analysis, emotion modelling, emotion mapping, and evaluation and iteration, and distil them into a theoretical workflow termed Semantic Closed-Loop Rapid Co-creation (SCRC) for organising diverse patterns of AD iteration. Our survey reveals three primary capability enhancements afforded by LLMs: semantic distillation, plug-and-play emotion perception, and real-time self-supervised evaluation. The LLM-centric design methodologies and analytical frameworks proposed herein provide theoretical underpinnings and practical references for the sustained evolution of Affective Design.