Discussion of all things patent mapping and analytics.

  • Home
    Home This is where you can find all the blog posts throughout the site.
  • Categories
    Categories Displays a list of categories from this blog.
  • Tags
    Tags Displays a list of tags that have been used in the blog.
  • Login
    Login Login form

Viva la Difference - how Cluster Searching differs from conventional searching (July 2015 update)

Posted by on in Amberblog
  • Font size: Larger Smaller
  • Hits: 3818
  • Subscribe to this entry
  • Print

This is an update of a blog first published in August 2014. Since then our terminology has changed from Automated Search Reports to online Cluster Searching but the results below would still apply. 



In recent blogs Ambercite has been promoting our newly developed 'Automated Search Reports'. (Now 'Cluster Searching', which will be the term used for the rest of this blog). We have also produced a number of reports for commercial clients, and speaking to a number of thought leaders in the world of patent searching and analysis about how the algorithms used in these reports differ from the processes used in other search processes. And in doing so, we hear this question a lot:

How do the results differ? Do you learn anything new with Cluster Searching? 

This led Tony Trippe of Patinformatics to propose a trial to directly compare the results of an Cluster Search to a conventional search. Ambercite is closely aligned with Australian IP law firm Griffith Hack, who run a comprehensive search department. So we asked a professional patent searcher to

  1. Take a patent that somebody might want to invalidate
  2. Run a conventional search process on this patent
  3. Report on the relevance of each patent found using this process

In the meantime, we ran one of our Cluster Searches on the same patent, and then the professional searcher would also rate these patents. In this case, the Cluster Searchi comprised of  the 10 most similar direct (known) backward citations combined with 59 not-cited prior art patents, or 69 patents in total (Categories D + E in the Trippe blog)

And what did we find? Tony's blog contains the full answers to this, but some key results are shown below.


How relevant were the patents found by these two techniques ?

Searcher assessment of relevance of patents found


Standard keyword search (595 patents found)


Cluster Search for 10 closest combined with 59 indirect prior art patents (69 patents in total)



47 (8%)

9 (13%)

Potentially relevant

128 (21%)

20 (29%)

Not relevant

418 (70%)

41 (59%)


So in other words, the 69 patents found by the Cluster Searching had a higher percentage of relevance (13% vs 8%) than the 595 patents found by the conventional patent search.

There is another term for "% relevance" that is sometimes used in statistics, and that is "precision". Or in other words, this trial showed that the 69 patents found in the Cluster Search had a higher precision than the 595 patents found in the conventional search. 


What was the overlap?

Just 9 patents were found in both the conventional patent search (595 patents in total)  and the Cluster Search (69 patent in total).

Both of the above results are shown in the image below.



Take home lesson

If you want to find all of the relevant prior art for a given patent, you should use both conventional patent searching and our Cluster Searching


Want to know more?

Ambercite offers free and completely confidential two week trials to Cluster Searching - please contact us to arrange a demonstration and trial. Testimonials for Cluster Searching are found here, including:

"Ambercite Clustering has managed to create a tool to enhance chemical patent due diligence in a remarkable way…revealing art which may cover compounds in discovery within claims of a generic scope.  Congratulations!" - Patent Analyst, global pharmaceutical company, USA

Rate this blog entry:

Mike first developed an interest in patent data when working as a research scientist, and deepened this interest when working as an IP manager which led to his role at Griffith Hack. Mike has published in the areas of chemical engineering, patent management, the value of patents and the use of patent data in in a wide range of publications and forums, including the international journals Les Nouvelles, and Managing Intellectual Property.