Machine Learning Meets Economic Geography: Alternative Data and Methods for Mapping and Analysing Geographies of Knowledge Production and Knowledge Relations

In economic geography, quantitative empirical studies have used mostly secondary data on patents, scientific publications, and R&D projects. These studies have contributed to the understanding of how knowledge production is geographically bounded, and of how regions, firms, and individuals create, maintain, and dissolve knowledge ties. However, more recently, scholars address the importance of using alternative data to address unresolved research questions (Duranton and Kerr, 2018; Fritsch et al., 2020).

The exponential growth of ‘big data’ coupled with enhanced computational capacity and high-performance machine learning techniques provide a range of new opportunities for mapping and analysing geography of knowledge production and knowledge relations. To name a few, this ranges from relational web data on firms in multiple countries (Abbasiharofteh et al., 2021; Kinne and Lenz, 2021) to Twitter data and news items (Ozgun and Broekel, 2021), but also digitized historical newspaper archives (Peris et al., 2021). Such data sources however require specific techniques for cleaning, manipulation, and analysis developed by the machine learning community. Whereas several scientific communities such as computational social science, networks science, and applied economics have started to take advantage of machine learning techniques (Emmert-Streib et al., 2020; Muscoloni et al., 2017; Storm et al., 2020), it seems that the economic geography community, to some extent, has not fully leverage the power of such methodological toolboxes.

We therefore invite contributions on methods of mining and analysing alternative data on geographies of knowledge production and knowledge flow. This includes, but not limited to, the following topics: