Keyword searching really is a flawed process...but we can do better
March 1 2018
Besides developing and marketing Ambercite I do work on patent searches myself, and a couple of recent searches have truly bought home the limitations of traditional searching.
Both searches were for what I would call mechanical inventions - a supposedly new combination of existing features that produced a claimed invention. The challenge was searching on these inventions. We were able to suggest some relevant keywords, but as was so often the case we knew that we would not get all of the right keywords, as different applicants use different terms for similar concepts. Or maybe they were emphasizing different elements of similar concepts.
More to the point, even with this incomplete combination of keywords we still ended up a with a long list of mostly irrelevant patents.
Did CPC or IPC codes help? To a point, yes, but it did not take to work out that the relevant inventions we did find were spread out among many different class codes. In one search I was able to use a combination of three patent classes as a filter - in the other search, this would been too risky.
Luckily, of course, we were able to use Ambercite AI as part of the search process.
Ambercite does have not keyword searching itself, but after a long and painful review of the patents we did find with keyword searching, we had found enough to form the basis of an Ambercite query.
This second stage of using Ambercite to look for similar patents to the patents we found in stage one was much more productive - to the point where 90% of the relevant patents we ended up reporting were found in the Ambercite search, particularly after a couple of iterations.
So why did Ambercite succeed where conventional searching was so inefficient? In Ambercite, we throw away the assumptions about what keywords and class codes are important - instead an AI algorithm analyses our database of 100 million patents and 150 million patent citations to suggest similar patents - regardless of language or class codes.
By doing so, it draws up on the collective intelligence of all of the patent examiners and applicants that have gone before. People smart enough to know, for ezample, that any of the following terms:
May all be different ways of expressing the same overall concept - depending on overall intent of the invention - and so identify connections between patents that the best semantic search engines would very likely miss.
And all of this is done in a patent search engine is both unusually easy to use, and shows results in a very to use review box.
Want to try Ambercite for yourself?
We have free trials and description presentations - please contact us to see Ambercite for yourself.