External collaboration and local knowledge divergence in regional knowledge spaces

This study investigates the association between regional collaboration structures and the semantic position of locally produced scientific knowledge. Drawing on the Regional Innovation Systems framework, it distinguishes between intra-regional and inter-regional collaboration as complementary dimensions of scientific knowledge production. Using over nine million Web of Science publications from European regions between 1998 and 2021, this study employs semantic text embeddings to measure local knowledge divergence, defined as the semantic distance between each publication and the established regional knowledge base. Within the same region–field–period, externally collaborated publication groups exhibit slightly higher average local knowledge divergence than internally collaborated groups. However, the estimated difference is substantively negligible and varies across periods. The principal pooled-sample finding is an inverted U-shaped association between external collaboration share and regional local knowledge divergence, with the fitted curve reaching its maximum at an external collaboration share of approximately 45%. The realised magnitude of this association is modest, and heterogeneity analyses show that it is most pronounced in life sciences and biomedicine and after 2010. These findings identify a context-dependent association rather than a causal or universally optimal collaboration structure, demonstrating how regional scientific collaboration portfolios relate to the evolution of publicly disclosed knowledge trajectories.