Non-toxic from the beginning- Analyzis of Patent Data

Important results from the project

The purpose of the project is to evaluate whether AI methods based on patent data can be applied in PRV and the Swedish Chemicals Agency´s activities. A concrete user case has been used where we have studied whether AI methods can be used that analyze patent data and identify potentially dangerous chemicals early in the product development process. The project shows that there are good conditions for continued cooperation between the authorities. However, both authorities need continued support from third parties for further development of AI technology in the field of patent data.

Expected long term effects

Patent data relating to a group of bisphenols has been used to train a set of AI methods to identify relevant patents by analyzing their text and image content. The study gives very good results when using text content. These results are already achieved by using relatively simple language technology. There is thus potential to apply language technology to patent data. AI methods trained on generic images are not applicable to identify relevant patents. These methods need to be adapted for the types of images that appear in patents.

Approach and implementation

Developments in language technology have made it possible to do semantic search in text, as opposed to traditional keyword-based search. AI can similarly also be used to represent content in images. We have evaluated the possibility of developing such semantic search tools to be able to automatically flag when chemicals begin to appear in new areas of use via patents that probably need to be examined in more detail. However, in order to cover several method families within AI, we have selected and evaluated both relatively simple methods and more advanced techniques.