Evaluating Interdisciplinary Knowledge Integration: A Coupling Approach Based on Concept Semantic Networks

Interdisciplinary knowledge integration fosters scientific innovation, yet existing metrics that rely on co-occurrence or citation proxies miss semantic relationships between knowledge units. We propose a novel concept semantic network coupling framework to fill this gap. Leveraging Large Language Models for precise entity extraction and relation annotation, we construct temporal knowledge networks from unstructured scientific texts and quantify interdisciplinary knowledge integration by edge-coupling, tracing directed knowledge flows via Time-Lagged Cross-Correlation. Applying this approach to Artificial Intelligence (AI) and Neuroscience (NS), we find that AI is predominantly method-driven, while NS is concept-anchored. We identify asymmetric knowledge flows: a concept-led flow from NS that shapes foundational AI methods, and a method-driven reverse flow from AI that accelerates NS discoveries. Our framework contributes a finer semantic granularity and high interpretability for interdisciplinary knowledge discovery.