Second PatentSemTech workshop at SIGIR'21: "In 20 years’ time, AI methods in patent analysis will be standard"

On 15 July 2021, the PatentSemTech'21 workshop took place as a one-day online event in conjunction with the International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR'21). The well-attended event with more than 40 participants from research and industry focused on the application of machine learning methods for patent analysis.

August 10, 2021 — The first keynote speaker was Noriko Kando (NTCIR), who spoke about the history and impact of evaluating patent-related IR (information retrieval) tasks in a series of workshops over the last 20 years. In the second keynote, Osmat Jefferson spoke about the role the lens.org platform plays in the innovation system, which is in constant flux. She pointed out the need to expand the FAIR approach to include the factor E (Enable), which is essential for "bridging the knowledge silos" and making patent information accessible and usable for everyone. Given the fact that the effective use of innovation is still very limited for a very large number of countries and target groups, a free and open platform such as Lens is of big importance for many groups.

This year's workshop focused on the application of new machine learning methods and their application to patent data. The audience had the opportunity to gain an insight into highly topical research results such as "Chemical Reaction Reference Resolution" and "Patent Text Generation with GPT-2". Within the last few years, more and more next-generation semantic technologies and machine learning methods such as domain-specific embedding models (PatentBert, SciBERT, etc.), deep learning (DL) and natural language processing (NLP) have been applied successfully to scientific texts, including patents.

In addition to scientific papers, various demos and case studies were presented in the workshop - for example, Siemens PatentExplorer or WIPOPearl from the World Intellectual Property Organization. Short presentations on topics related to the application of machine learning methods and semantic technology addressed, for example, Linked Open Data (LOD) in the context of special use cases in industry and publicly funded projects such as IPLodB.

In the joint expert panel, scientists and participants from industry discussed the question: “AI and Patent Analysis - Friends or Foes?” Their unanimous conclusion was that in 20 years’ time the application of AI methods will be standard in the field of patent analysis, too – despite the resistance of many stakeholders. At the same time, the importance of high-quality datasets for training machine learning methods was pointed out. The necessary adaptation of AI methods to the workflows under consideration, e.g., in the context of patent application procedures, was also emphasized.

The presentation slides and further materials on the PatentSemTech’21 workshop will soon be available on the workshop website.

The proceedings have already been published as Open Access: http://ceur-ws.org/Vol-2909/

In addition, two of the best papers have been proposed for publication in a Virtual Special Issue (VSI) of the World Patent Information (WPI) Journal.

The PatentSemTech workshop series is to be continued next year as part of SIGIR. In addition, the organisers are considering the possibility of focused workshops within the framework of other conferences such as ISWC or Semantics.

Contact:
Dr. Hidir Aras, Project Leader TDM, IT Development and Applied Research
hidir.aras [at] fiz-karlsruhe.de (hidir[dot]aras[at]fiz-karlsruhe[dot]de)
https://www.fiz-karlsruhe.de/en/forschung/text-und-data-mining-tdm