How to Find New Industry Partners for Public Research: A Classification Approach

Karl Trela, Yuri Campbell, Friedrich Dornbusch, Anna Pohle
2020 IEEE transactions on engineering management  
Finding new industry partners poses a challenge to many public research organizations. This article explores how statistical classification can support partner selection at the example of the Fraunhofer Society in Germany, Europe's largest public organization for applied research. We use internal cooperation data and feature sets based on unstructured data, i.e., text and industry codes, both of which describe business activities of firms. An important advantage of this data is that it is
more » ... ble for most companies in Germany, even small and medium enterprises, which allows for an almost complete screening of the market, in contrast to using other data sources, e.g., patents. In addition, we also include economic variables linked to firms, as turnover, number of employees/managers and firm age. We report the performance of various classification techniques such as logistic regression, support vector machines, and random forests in our dataset for diverse combinations of feature sets. Results show that simple methods with fewer parameters remain competitive in comparison to complex ones. Overall, the performance of most classifiers is high enough to support the decision process of finding new industry partners for public research. Index Terms-Collaborative research, knowledge and technology transfer, machine learning, natural language processing, partner selection, text mining, university-industry cooperation.
doi:10.1109/tem.2020.2992060 fatcat:ufm2ucx3rne2pand7644aah5za