Tracing Cognitive Dynamics in Group Ideation: An LLM-Based Framework for Automating Linkography

Collaborative ideation is a cornerstone of creative design, where the evolution of ideas reflects underlying cognitive processes such as association, integration, and refinement. To understand these processes, researchers use linkography, a method that maps semantic links between ideas, but it relies on manual coding and limits reproducibility and broader application. This study proposes a linkography framework that decomposes the construction process into binary classification tasks and uses Large Language Models (LLMs) to automatically identify new, supplementary, and modified ideas, generate the linkography, and compute creativity-related metrics such as link ratio and fluency. We evaluated the accuracy and validity of the framework on 26 group brainstorming dialogues (including 3,553 utterances). The framework demonstrated high consistency with human annotations across Fleiss’s Kappa, accuracy, F1, and IoU metrics, outperforming a BERT-based baseline for link annotations. Its creative metrics also significantly correlated with human-evaluated creative performance. This study provides a consistent and scalable methodological framework for studying human cognitive dynamics in creative design. It supports both research on collaborative creativity and the future design of AI systems for human-AI co-creation.