Essays on Applications of Machine Learning to Science, Patent, and Economic Data

This dissertation studies knowledge flows with the help of advanced machine learning models and analyzes the impact of the new European Unitary Patent system on patenting decisionmaking within organizations.

The first essay asks whether we can detect research projects that go against established wisdom, question traditional results, and use novel approaches. We analyze grant application data from the Volkswagen Foundation, Germany’s largest private research funding organization. Our approach tries to compare “unorthodox” grant applications with the spectrum of existing scientific literature. A key tool in our analysis is Logic Mill, a knowledge navigation system developed during this dissertation. This scalable and openly accessible software system identifies semantically similar patents and scientific publications using advanced Natural Language Processing.

The second essay addresses the question of whether we can detect and trace knowledge flows from scientific publications to patents and vice versa. Our approach uses specialized machine learning models that can indicate similarities in patent-paper pairs. These models are trained and evaluated on customcrafted datasets.

The third essay analyzes the impact of the newly created Unitary Patent system on the decision-making of patenting organizations. This new European patent system is a historic step for innovation as it gives patent owners access to centralized enforcement, albeit with the risk of centralized invalidation.