Data analysis of patent information using AI

Patents contain information about the use of chemical substances in products or articles at a stage before they have been introduced to the market. Early detection of patents linked to the use of potentially hazardous chemical substances and high exposure applications would open the door to proactive approaches to prevent these chemicals from entering our cycles. Such approaches will require AI, since traditional keyword-based searches are insufficient when both substances and applications are often new and not known in advance. 

In this pilot study, SCA and PRV wanted to explore the possibilities of identifying relevant patents using AI and recruited RISE as the executing partner for this work. Patents related to a group of hazardous chemical substances (bisphenols) was collected from the European Patent Office (EPO) and this dataset was then used to train a set of AI methods to identify relevant patents by analysing their textual and image content.

The case study shows surprisingly good results when using textual content. When relevant patents are compared to randomly selected patents, an AI method can identify 96% of all relevant patents while 91% of the patents identified as relevant are also relevant. These results are already achieved using relatively simple language technology. The project thus shows that there is great potential for applying language technology to patent searching.