The Inventive Step was founded in the beginning of 2020 as a result of a thesis research, executed by one of the founders at the Intellectual Property & Standards department of Philips. During this project, research was executed into the quality and usability of automated patent mapping for IP analytics. Automated patent mapping refers to tools capable of text mining for technology mapping. The output of these mapping tools represent the technological content of a certain set of patents in a map that borrows its appearance from cartography. With these technologies, huge amounts of patent records (up to millions) can be analysed and clustered based on their technological content, and can be represented in an easy to interpret manner. Typically, these automated technologies are found in the premium licences of IP analysis software.
The results of this research showed that the tools are highly efficient. However, the current state-of-the-art in automated mapping also demonstrated some severe technical limitations: the accuracy performance is limited and fluctuating, the mapping output is too susceptible to inconsistency and the descriptive labels for technology fields are often vague or meaningless. Research was also executed into the technologies and design choices within the software of the tools (reverse engineering). It appeared that there is a significant need for improvement in these automated technologies in order to transfer attractive visualisations to true valuable intelligence.
As artificial intelligence and more specifically natural language processing (NLP), has evolved over the years, we identified a huge potential by integrating and tailoring these techniques for IP analytics. The Inventive Step distinguishes itself from the installed base by not offering our clients an extensive IP-analytics package with multiple applications. We purely focus on high quality content analysis. By implementing advanced forms of NLP we take work out of the hands of the Inventor, Searcher, Analyst or Attorney when it comes to content based exploration. Automated technologies are capable of analysing vast amounts of data in a fraction of the time it would take a human being. By doing this, we are able to make vast amounts of patent content data easily accessible to the world. The scope of potential end-users is therefore not limited to IP experts.
Our technology is focussed on finding relationships between patents based on the descriptive meaning of words using actual linguistic interpretation. This more closely simulates how an actual human would analyse a patent. Combining this analysis with metadata such as CPC codes, families and forward and backward citations, we foresee both high recall and precision. We train deep learning models for NLP, which are able to predict relationships between words in a certain sentence. These relationships are used to extract the key concepts and the uniqueness of a patent. Next to that we also focus on a consistent output of our analysis. In short, we apply advanced NLP to patent content in order to find relationships between documents based on their technical content.
The high potential there is for NLP in patent analytics, became clear after a study into the state of the art in this fast evolving field of artificial intelligence. Literature and empirical results of applications in other fields such as marketing and market intelligence, machine translations and speech recognition shows the potential NLP beholds and the impact it will have on all of our lives. After this exploratory study, we got in contact with several potential end-users of our technology: Patent Attorneys, IP-analysts, Patent Searchers, IP-lawyers and R&D parties. As we believe that a clear view of the needs of the end-user in an early stage will increase quality and acceptance in the future. Together with these stakeholders we identified multiple scenarios in which our techniques could be beneficial and valuable for the user. These scenarios, or use-cases, can be translated into concepts of how the tool should function. We selected one concept to develop into a prototype with which we are going to perform test cases together with stakeholders. These test cases will function as a proof of concept. The prototype is developed to find related documents to a certain invention, which can for example be used for: novelty searches, freedom to operate searches and technology exploration for R&D and scholarly use.
With our technology, we deliver tools capable of analysing large amounts of technical data in a complex environment like the one of intellectual property, to present the data to the user in a comprehensive manner. This increases the accessibility to techniques and insights conserved in patents. The tool will then empower the constitutional goal of the patent-system: making technology available to the public in order to create value for society, in exchange for a temporary monopoly.
In our team we combine the essential forces to integrate advanced NLP into Patent analytics. The Inventive Step is a part-time endeavour arised from a shared passion for automation and scientific computing. All team members have full time jobs next to this project. Co-founder Jaap van de Vorst holds a degree in Technology Management and wrote his master thesis at the IP-Analysis department at Philips about automated technology mapping. He is currently working as a System Architect at a Dutch technology start-up. Co-founder Rogier Versteeg, holds a degree in Aerospace engineering and wrote his master thesis on implementing state-of-the-art deep learning models in Aerospace applications. He recently started working as an Application Developer, where he develops essential skills for building the front- and backend of our tool.