Go to https://app.cipher.ai/custom-classifiers/home 

NOTE: Access to Cipher Classifiers is part of your account settings. Please contact your Cipher Account Manager or email training@aistemos.com to request access.


1. Click ‘Create new Classifier’. 

2. Name your Classifier e.g. Electric toothbrushes

3. Include scope notes as a helpful reminder of what this technology is e.g. Motorised oral hygiene cleaning appliance

4. You’re now ready to start building:


The classifier needs a base understanding, by way of examples of patents, of the boundaries of your search. This needs to include the following:

a) Technology you are interested in, 20-50+ patents are recommended (Positive)

b) Technology you are NOT interested in, 50-100+ patents are recommended (Negative)

More explanation on the Training Set can be found here

To get started:

If you already have a list of these to hand (e.g. a list from a previous search), you can upload the examples by clicking ‘Options’ > ‘Add Positive’ or ‘Add Negative’. 

Global search tab:
If you want to run a quick search to help you identify some positive and negative examples, enter a query in the search bar of the Global search tab. 

Review the list of search results and mark the patent as either Positive or Negative. 

The image below illustrates a list of search results for the query “electric tooth*” and where the results have been marked as Positive, Negative or Ignored if unsure. 

The count in the top left will populate to show you how many patents you have marked as either Positive or Negative (your Training Set): 

Suggestions tab:
This will begin to populate with a list of patent suggestions that Cipher thinks you might be interested in, once you have started to identify positive examples. These are patents that are not (yet) part of your Training Set. 

5. Build the Classifier: Once you have at least 5x examples of Positives and Negatives in your Training Set, you can click ‘Build classifier’. 

This will assign a Score to the list of patents you see to continue to classify. 

Score = what percentage the classifier has assigned to the patent as being related to the training set. This will change the more you train and recalibrate the classifier. 

TIP: use the Global Search and Suggestions tabs in conjunction to review the Score, and help you to feed in further Positive and Negative examples. I.e. if you continue to see patents that mention e.g. hand-held motor, you can run a Global Search on hand-held motors and mark patents as negative so the classifier knows that this is what you are NOT interested in. 

Note: If you mark something as “negative” which the Classifier believes you have marked something similar as “positive” previously, it will continue to question you on this technology to ensure it’s understanding of what you are interested in is correct.

As you continue to train and rebuild the Classifier, a strength indicator bar will inform you how much training is still required.

6. Evaluate tab: these are the datasets you want to search using the Classifier you’ve built. E.g. companies of interest, or a Global search, or an uploaded list of patents.

This view provides a quick view on the Score of relevancy of patents within the datasets of interest, allowing you to establish if you want to continue training the Classifier or not. This does not show a Score of ALL patents (unless ALL is selected) - just those that Cipher finds to be either Positive or ‘Borderline’. 

At this point you can:

a) Download the results to see the Score assigned to every patent in your evaluation dataset

b) Create a Cipher report for analytics 

c) Return to the Global Search and Suggestions tabs to continue training your Classifier.

Did this answer your question?