Dear Colleagues,

Our article accompanying the PIUG talk is now available at Deep Learning AI for Patent Search and Analytics, along with links to additional material.

Thank you for your questions and comments at PIUG - we hope they were answered. We welcome any lingering questions.

We invite you to participate in our demos and trials, and help us fathom "deep" learning and secret patent "pass-phrases" (smile) 

best regards,

Sumeet Sandhu 

  • No labels


  1. Hi Sumeet,

    It was great seeing you again last week. I'm wondering whether you have given consideration to incorporating existing controlled vocabularies into your deep machine learning, as it benefits from human analysis and interpretation of abstracts and full text to consistently index concepts which are expressed in a variety of ways in the text. I'll be happy to discuss with you further.



    1. Sumeet Sandhu AUTHOR

      Hi Elliott,

      Thank you for the comment.

      Automatic vocabularies from deep learning do need curation - even if we were to separate out types of 'onyms' in future versions of deep learning, there is usually a specific nuance the searcher is most interested in. Automatic vocabularies can be curated with a feedback loop on a per search or per project basis.

      If there exist pre-curated vocabularies drawn from patents, they can only help, and may even speed up the process. It would be great to discuss further - please feel free to email

      best regards,


  2. I have just finished reading this paper and whilst this is interesting and seems to show an improvement over early AI attempts I believe that this still ignores one major factor.

    How does this solution compare to the professional searcher?  I'm not an electrical engineer or expert in phone technology but I doubt very much if I would have chosen to conduct a "professional" search using either Google or Innography.

    As a busy searcher working for a large FMCG company I don't have the time or the expertise to conduct a "professional" comparison but I wondered if a retired searcher (or outsourcing search company) should rise to the challenge.

    Does this matter?   Well if your company has to make a business decision based on 3rd party IP rights you need to have the best result possible - not one that'll do.  If a company relies on AI tools when a patent searcher would have done a better job does this mean that your potential liability is increased by just using tools?  As professionals we need to provide the best advice.

    I'm not deliberately rubbishing the developments in AI tools I just want a reality check.

    Best regards

    Nigel (searching chemistry since 1979)


    1. Sumeet Sandhu AUTHOR

      Hi Nigel,

      Thank you for your comment. We welcome comparisons to other solutions.

      We believe the new AI provides a more efficient tool for the professional searcher, as well as a learning tool for the casual searcher. The low hanging fruit is - reduce the cognitive burden of simpler tasks like looking up synonyms, and crafting queries iteratively. 

      In fact, we would very much welcome an evaluation of search in classes like :

      A61K38 - Medicinal preparations containing peptides
      A61K39 - Medicinal preparations containing antigens or antibodies

      best regards,