Trace of Knowledge: Benchmarking Novel Text Mining Based Measurements

The impact of public research outcomes on economies, and societies, in particular, in terms of innovation and development is widely accepted and empirically investigated [9, 3]. However, many studies suggest a systematic underestimation of the impact and benefits of public research. Empirical studies describe that current approaches capture only specific aspects of knowledge transfer between public research institutions and private entities. The main interrelated reasons contributing to this systematic underestimation are that most established knowledge transfer measurements focus on intermediaries and use proxyindicators like patents, licenses, spin-outs and co-publications as data sources, but these
metrics are problematic because they can result in type I and type II errors, since many of them capture a transfer that is never utilized by a private entity (e.g. like unused patents). In addition, there are occasions where the proxy is not met so the actual use is not being captured.

We try to improve this systematic underestimation by adapting novel computer linguistics methods to this field and putting them into perspective with the existing measures of knowledge transfer. We use both basic and more advanced statistical learning tools from the field of computational linguistics and statistical learning to trace the knowledge fragments[2, 6]. In addition, we utilize a mixture of standard algebraic and probabilistic methods. Furthermore, pattern recognition, classification algorithms help to trace the public research outcomes, going beyond plain word co-occurrence.