Examination of overlapping boundaries of innovation systems using deep neural network and natural language processing

Systems of innovation (SI) studies previously considered innovation at different geographical levels and with different foci on actors and processes. However, the boundaries of the SI concepts are becoming blurry and shifting due to technological and economic factors, globalization and interdependence between countries and firms. Previous literature has limitations to address shifting boundaries by assessing SI literature holistically. This study set out to understand the overlaps of SI concepts. It aims at delimiting the SI field, delineating boundaries and at determining the topics underlying the SI and overlaps. Context-specific synergies become transparent, enabling decision-makers and researchers to consider economic interrelationships in a topic-specific and cross-system perspective. The study uses deep learning in combination with natural language processing on data consisting of 2380 SI literature related article abstracts retrieved from Web of Science and EBSCO research databases to capture the semantic context of topics, whereas external information is used to cross-validate interrelationships and to observe topic distribution. The findings reveal significant proportions of overlaps between SI apart from regional and technological systems. Various topics that are discussed within the SI literature such as industrial clusters and structures is a common topic for all SI. This illustrates that economic topics are contextually or theoretically not bound to a system but distributed across the entire literature. We therefore provide a structured framework to illustrate SI overlaps and to address topic synergies. Our study contributes to the body of literature where SI boundaries are investigated specifically to the determination of system overlaps, reasoning of their identical SI characteristics and examination of cross-system thematic synergies.