Have you ever desperately tried to find an item, which you without doubt remember being in certain type of container, say, a blue box? You repeatedly scan each corner, closet and locker for the blue box, but after a good while you must admit: it’s not there. Sitting down and letting the frustration settle, you come up with a radical idea: what if I put it in that green bag instead of a box… After redefining your search target, it takes only a second to find the item. You saw many times what you are looking for, but did not recognise it.
There is an analogy to patent searching: if the search target is wrong or vague, finding what you need takes a lifetime.
The usual way of doing a patent search is to log in to a patent search service, type in a few keywords, be overwhelmed by the thousands of hits produced, add some filters, and start crawling through the results. And to iterate with different keywords and filters when frustrated. At least the haystack is there. But am I looking for a needle or what?
Another way is to use a so-called semantic search or similarity search. Copy-paste a long invention disclosure form into a web form and click search. That’s easy. Again a bunch of results, sorted by relevance. But relevance to what? What did the machine understand from my IDF, which is like pincushion full of needles?
In the patent world, it is always a matter of technical relationships between things. Both internally between key features of your invention and between the features of the invention and prior art. Managing these is not an easy task for a human, let alone a computer. However, a computer learning from the best human professionals can be an unbeatable companion. Defining the user's search target better and training the search engine with previous patents and their novelty search results with modern machine learning techniques helps to solve that equation.
The search target is the reference point, defining what one want to find, whereas the search case determines the data space within which to search and the logic of the search. All these can be fundamentally different for different purposes.
For example, in a novelty search, the search target is your invention compared with every public document available. In an invalidation search it is your competitor’s invention compared with documents before the priority date of the competitor’s patent. Only a document that discloses the same or more specific features in a similar technical relation to each other as in your or your competitor’s invention, qualifies as a novelty bar. Therefore, knowing what is essential to the invention and the whole content of prior art, is crucial.
In a freedom-to-operate search, the search target is your product compared with claims of granted patents that are still in force. What you need (but usually don’t want) to find are patents whose claims are broader than your product’s particulars. This is very different search case and usually even more laborious to conduct with traditional search tools than novelty searches.
In the future, we will see a growing number of search products that are designed for very specific tasks, but tuned to the peak in these.