Exploring disruptive innovation through deep learning and patent analysis
This project aims explore and predict potential disruptive innovation opportunity. The disruptive innovation opportunity means potential and promising technology and innovation to penetrate and reshape mainstream as well as niche market. In other words, we find answer the questions – first, which technology will penetrate new market? Second, how is this disruptive technology implemented as business?
The main technique to solve these questions are machine learning and deep learning. Deep learning is a technique for learning and inference based on big data and it can solve hard problem in practice without information loss. Especially, it provides ‘right’ insights through iterative learning with high performance improvement. Machine learning and deep learning support are well-suited to deal with various type of data – both structured and unstructured data as coming of big data age.
This project serves as an ex-ante forecasting while considering existing technological paradigm because we will use both unsupervised learning and deep learning such as semi-supervised learning. It enables to early detection and forecasting by accumulating newly generated data not evaluating and classifying after implementing technology.
In addition, this research examines where innovation trajectories begin and whether there is such a thing as a “patent zero” that triggers a new technological trajectory. It focuses on understanding the early stages of technology development through patent analysis and investigates whether disruptive or breakthrough innovation pathways can be traced back to specific trajectory-initiating patents. Building on the concept of technological trajectories as directions of advance within a technological paradigm, the research explores how technologies evolve, converge, and diverge over time.
A particular contribution of this work is the refinement and analysis of the concept of patent zero, defined as the initial patent that marks the beginning of a new patenting trajectory. The research investigates the characteristics of such patents, including limited prior art, broad and inventive claims, high impact, and large patent families, and examines how these properties may be translated into patent indicators. It further asks whether patent zeros actually exist, in what forms they appear, and how they may be systematically identified across different technological fields.
Furthermore, a five-phase framework for disruptive innovation pathways has been developed to explore how these phases can be operationalised using patent data. Based on Christensen’s theory of disruptive innovation, this work proposes an early warning system for identifying potentially disruptive innovations by combining theoretical insights with patent indicators such as originality, radicalness, impact, scientific base, and technological disruptiveness. In this context, patent zero is introduced as a possible starting point for disruptive innovation pathways and as a foundation for future automated identification methods.
Overall, the project extends beyond deep learning-based forecasting by also investigating the origins, pathways, and early warning signals of disruptive technologies through patent analysis. Its long-term goal is to automate the identification of disruptive innovations and patent zeros in order to support researchers, managers, and policymakers in understanding and responding to emerging technological change at an earlier stage.
Project team: Dr Melanie Martini, Dr Leonidas Aristodemou, Dr Marcus John, Prof Sven Schimpf, Prof Frank Tietze, Dr Yujin Jeong
Related publications
Melanie Martini, Frank Tietze, Marcus John, Leonidas Aristodemou, Alexander Schönmann, Sven Schimpf (2025). Where do innovation trajectories start? Is there such a thing as a “patent zero” that triggers a new trajectory? EPIP Annual Conference 2025.
Martini, M., Tietze, F., John, M., Aristodemou, L., Schönmann, A., & Schimpf, S. (2024). Conceptualizing Disruptive Innovation Paths, Patent Zero and Patent-Data Based Operationalization. XXXV ISPIM Innovation Conference 2024. https://doi.org/10.24406/publica-3571
Yujin Jeong, Leonidas Aristodemou, Frank Tietze (2019). Exploring disruptive innovation opportunity using deep learning. The R&D Management 2019 Conference, Paris, France.