Measuring innovation: computational text analysis and machine learning applied to start-ups, entrepreneurial ecosystems and European cooperation

This thesis develops new approaches to measuring innovation through computational text analysis and machine learning applied to textual and numerical databases. The first chapter introduces a dual framework for measuring the novelty of start-ups’ ideas and inventions (LDA, Kullback-Leibler), demonstrating that novelty does not function as a uniformly positive signal. The second chapter compares the determinants of fundraising across three ecosystems (US, EU, MENA) using Random Forest and SHAP, introducing the concept of investment identity. The third chapter assesses the impact of the COST Action (Horizon 2020) using a mixed-methods approach (bibliometrics, NLP scoring, interviews), demonstrating that the impact operates through the formation of networks. The thesis contributes to innovation management and marketing strategy at the theoretical level (dual novelty, signal congruence, investment identity), the methodological level (KLD and SHAP in management sciences) and the managerial level (calibration of start-ups’ competitive positioning by ecosystem).