Most people when starting a patent search start by looking for patents with similar keywords, in either one of the free patent search engines (Google Patent, Espacenet, The Lens or others) or one of the subscription engines out there. And indeed this is a good approach and one that many users of Cluster Searching also use.
Similarly, a lot of people combine this with class code searching. Class (or classification) codes can be useful means of conducting a preliminary filtering, or even a search in some cases.
Nonetheless, in many cases these two approaches can be shown to be insufficient in themselves:
- The biggest reason is the inconsistent use of technical terminology. We have published a number of case studies illustrating this – for example this one for a pilot monitoring headset. In some cases it is possible to predict the most likely synonyms for technical terms, but in other terms it is either practically impossible, or including every technical term would blow the result set up so that the number of patents would be impossible to review in a timely manner
- Similarly class codes can be inconsistently used – the most similar patents can have very different class codes to what you were expecting.
So what can you do to avoid these risks? Ambercite developed Cluster Searching to get around these limitations, and this applies advanced analytics on citation data to predict similar patents even if they have different keywords or class codes. We have a lot of case studies in our blog to show that it can work very well.
But it is worth asking – why does it work? There are two good reasons for this:
1) The value of collective wisdom
The main reason why it works so well that it combines the citation opinions of many different applicants and examiners. Each list of prior art citations for a patent in our network is an opinion by an examiner, and in many cases an applicant, of what earlier patents are similar and relevant.
Let’s think about some more. Each examiner and/or applicant is human – they may be brilliant, just average or somewhere in-between. When tasked with the job of finding prior art, they will form an opinion of what aspects of the invention are important, what keywords or class codes should be searched, and then run a search and select their short list of most relevant citations. Logically we would expect this list of citations to be mostly correct in terms of relevance, but maybe not 100% complete – because no patent search is 100% complete.
So that is one set of search opinions (or two if the applicant and examiner both list citations).
But this is only one data point in our network of patents, with an example of such a network shown below:
The next patent in the network is a new seat of patent citations – almost certainly listed by a second examiner and often by a second applicant, and therefore independent from the first set of citations. And if the first examiner or applicant was less than perfect (and we are all less than perfect) we would expect that the second search opinion would produce different results – again not perfect, but different.
And again for the third patent, and so on. And so over time, even if each individual patent search and searcher is imperfect, what should be a reasonably complete set of relevant patents will emerge.
Effectively this is the concept of ‘collective wisdom’, which is the well known princple that the collective is often brighter than any one individual, a well accepted concept that underpins teamwork in the corporate world.
2) Because direct citations can lead to new ‘unknown’ citations
One of my favorite real world case studies was a invalidation search for a granted US patent, which included the keyword ‘carbon dioxide’. Eventually using Ambercite tools we were able to find a new and highly relevant prior art document, filed by a Japanese company, used to the obscure term ‘carbonic acid gas’ instead of the common synonyms ‘carbon dioxide’ or ‘CO2’.
It is worthwhile thinking of why this patent was both missed – and found:
- It was almost certainly missed because nobody thought of searching for the term ‘carbonic acid gas’ instead carbon dioxide or CO2. No surprises there.
- It was most likely found not because a ‘super searcher’ recognised that they should also search for patents containing ‘carbonic acid gas’ – but instead because the USPTO examiner for the Japanese owned patent would have recognised that the likely prior art patents would have also used the more common terms ‘carbon dioxide’ or ‘CO2’, and so ran a search for prior art patents that listed these terms, as well as for ‘carbonic acid gas’.
These citation links would have created links in our network – which would have caused our network to suggest a link between the patent we were trying to invalidate, and the earlier Japanese patent. And so by doing so, overcome the issue that it would be unlikely that somebody would have searched for carbonic acid gas patents as prior art for patents referring to carbon dioxide.
And this way, relevant patents can easily be found
This principle is shown below. You can see how indirect links from between two patents, even if with different terminology, can be recognised by Cluster Searching if they in turn cite similar prior art.
Do you want to find patents missed by conventional searching?
If so, please contact us and we will be happy to set up you with trial access. We also offer free sample seaches for patent litigators looking to invalidate problem patents on behalf of their clients or companies.