Guided Innovation Feature Miner: A Design Science Artifact for Studying Innovation Diffusion with Deep Learning

There is a lack of concise design knowledge for extracting innovation features that characterize diffusion levels while providing actionable managerial insights. Against this background, two approaches currently exist in the literature: On the one hand, studies applying diffusion theory provide interpretable, theory-aligned insights but neglect predictive accuracy. On the other hand, machine learning-driven approaches optimize predictive performance but disregard theoretical coherence and managerial interpretability. Consequently, there is a need for systematically derived design knowledge guiding researchers and practitioners in building theory-grounded, yet predictively powerful innovation analysis tools. Responding to that need, this article presents a design science artifact, Guided Innovation Feature Miner (GIFM). GIFM integrates Rogers' diffusion of innovations theory with graph convolutional networks and hierarchical attention mechanisms to extract theory-aligned features from unstructured patent texts. Evaluation across 31,804 cybersecurity patents demonstrates that complexity (simplicity) and observability contribute most to diffusion prediction, while compatibility has the least influence.