Current research projects

The project will create a public database linking patents to scientific publications, using a high-performing text mining method to extract patent in-text references. As a result, it will make it possible for researchers to analyse the impact of scientific research on industry innovation.

Using a unique dataset of grant applications to the Research Council of Norway across all academic fields, the project will track patented innovation linked to the research grants, and derive implications on how the design of research funding can be improved to increase its impact on innovation.

Project stopped

Using a novel dataset linking declared-SEPs to different sources of government contribution, this project will assess the importance of government-sponsored research for SEPs and its impact on the development of technical standards.

The project aims at identifying specific characteristics in scientific clusters that signal their later impact on technological developments. For that purpose, it will: 1) use frontier quantitative techniques like semantic analysis and graph embeddings, applied to scientific publications and patent data; 2) conduct monographs on specific cases in various scientific fields (e.g. machine learning, cancer, mRNA, cybersecurity, CRISPR, quantum computing); and 3) measure and test specific characteristics of scientific discoveries that might influence their later impact. The deliverables will include an analytical report and a toolkit allowing users to assess the potential impact of specific scientific publications on technology.

Citations from patents to scientific publications are usually interpreted as indicators of potential knowledge flows. But citations are sometimes problematic for various reasons, while text-based alternatives suffer from loss of information and limited scalability. We propose, therefore, to harness the semantic similarity between patents and scientific publications using the latest advances in machine learning. Our solution utilises transformer models that identify semantically similar documents. A patent that is highly similar to a prior scientific publication may have been influenced by it. Our approach is scalable and can handle large amounts of text. Once documents are semantically linked, we use the data to draw inferences regarding the diffusion of knowledge from science within publications and to and within patents.

Current retrieval systems for patent search mainly rely on textual content to find similar documents. Research on how to use visual content in figures, such as diagrams and schematic drawings, or even a combination of text and images in patents ("multimodal search"), is still limited. However, a genuinely multimodal approach now promises to alleviate known issues in patent search, such as those introduced by the ever-changing terminology of technological development. The proposed project explores and extends state-of-the-art technologies for information extraction and similarity search in patent databases, focusing on deep-learning techniques utilising large-scale datasets. The overall objective is to exploit image information in patents and connect it with textual content for optimised retrieval.

There are many technology-based systems available to assist students. Some of those systems aid students during the learning process by proposing learning resources or recommending activities. Other systems support students in the assessment phase by giving feedback, or monitor their progress by recommending the best learning path to complete the course successfully. Depending on the students and course needs, some features are more suitable than others. This project aims at:

• defining a smart learning system to help students succeed in their learning process using artificial intelligence techniques for IP training
• defining the best automatic assessment strategy and learning resources for an automatic feedback system
• building a chatbot to simulate lecturer/student interaction.

Based on electronic doctoral theses (EDT) repositories, we match information on the doctoral graduates of France, Germany, the Netherlands and Spain to patent and scientific publication data. We thus identify both the graduates who become inventors and those who become scientists and publish articles cited in the patent literature, either directly (patent->graduate's publication citation) or indirectly (patent->other publications->graduate's publication citation chain). We collect the same information for the graduates' supervisors. We then explore the factors affecting the probability of graduates engaging in inventor careers and/or scientific careers, but with invention-relevant publications. These include gender and the supervisors' scientific and inventive records. For France and Germany, we also focus on the specific contributions of foreign graduates.