by Joni Sayeler/ Sep 30, 2021
Should you use AI-based patent search tools? We performed tests to find out how to get the most out of AI for patent searches
To search patents with AI or not?
Patent searching has been around for a long time, during which there has been a revolution whereby the patents were transformed from paper in libraries into data searchable on-line. One challenge this change brought was to advance from searching not only patent classes but all the accessible data in different fields. User interfaces and search formats improved, continuously assisting us patent searchers in handling the increasing amount of patent publications. Now we are in a new revolution going from a controlled search environment where we are using Boolean operators and data fields into AI searching where the AI defines the search environment.
As a patent information specialist, I love to control and steer my search (we are not control freaks 😊), and at the same time I love innovation, and this creates a conflict in me when it comes to AI patent searching. To minimize this internal conflict and frustration within me I want to have confirmation of how well the AI performs. So, I digged into literature on how to test AI and realized that I would need time and data. Then I started to think about other search engines such as Google, did I test them, or did I accept that they fulfilled their purpose? So my approach to AI patent search engines was, do they fulfil a purpose for me?
“Does AI patent search fulfil a purpose for me”
I started experimenting with AI patent searches using IPscreener, to see if they could replace my Boolean searches using Orbit, the answer to that was NO. The AI search did not replace me (Woohoo)!
Then I started testing with the AI search in the beginning, middle or end of the process just to learn where it provided me with the highest value. I learned that the AI search worked best for me in the starting phase of the search giving me a kick start for the search. The reasons are that I found it cumbersome and difficult to interrupt a flow in my Boolean searches to, all of a sudden, perform an AI based search. I had difficulties getting back to the search flow and felt distracted even though I got some additional good documents from the AI search. For the searches performed in the end it felt more like a check on my Boolean search and did not provide me with the increase in efficiency I expected. At first, my approach was to begin my search with an AI search, export the documents to the database I use for Boolean searches, and then review the documents with all the tools I am used to, such as highlighting, etc. But I found that the reviewing of the documents was taking more than double the time of a normal review. By jumping straight into the AI patent results, neither my eyes nor my brain knew what they were looking for and I was having to read each document more thoroughly, rather than scanning the documents as I am used to. I realized that by first doing a “Bullseye search”, which means a really narrow key word search, I could have time to think about synonyms for searching and highlighting. This gave my brain the information necessary for understanding the search so that I could scan documents instead of reading.
My conclusion is that AI patent searches fulfils a purpose for me by providing a quick start and is another tool I can use to provide my clients with more efficient and high-quality searches. The secret for me was to perform a search first to actively engage in finding and highlighting of key words to in order to effectively assess the results from the AI search. It was an awakening to realise that when I engage in a search then my eyes and brain trained to power read without me being conscious about it. And this does not come automatically when solely depending on an AI search.
There are several AI or semantic based patent search engines out there and I used IPscreener which worked well for me and led me to my conclusions. For more information about IPscreener visit https://www.ipscreener.com/