Introducing a novel AI-based text mining method illustrated through an analysis of German innovation proposals

In this study, we examine how standard AI tools such as ChatGPT can be used for the structured analysis of large text corpora. To this end, we analysed 482 applications from a specific innovation funding program of the German Federal Ministry of Science using ChatGPT. Thanks to ChatGPT's ability to cluster projects based on their characteristics, complex data sets can be systematically explored and patterns recognized that would have remained hidden in a manual analysis. It turns out that cluster formation controlled in advance by the user via cluster definitions (using prompts), is in some cases more meaningful than the fully automated cluster formation of tools such as BERTopic. The analysis of the 482 funding applications provides detailed insights into the state of innovation in Germany: 83 % of the proposals dealt with topics related to digitalization and social innovation (half each), while the remaining 17 % dealt with sustainability issues. While 77 % of all project activities focus solely on the early concept phases, only 17 % of activities relate to the piloting and validation of applied ideas. Correlation analyses examine the relationships and potential connections between the clusters identified in different categories, in order to uncover patterns and dependencies in the innovation application data. For example, the correlation data can be used to determine the “age” of certain fields of innovation. The study also demonstrates the suitability of the method for classifications with external cluster definitions such as the UN Sustainable Development Goals (SDGs) or the EU program “Horizon Europe” to assess the suitability of research projects, with regard to specific frameworks. This could be particularly useful for scientific funding organizations.