Training using Search (Explore)

User permission required: 'View sources' AND 'Review and label'.

Even if training using Search is not one of the main steps outlined in the Explore phase of the training, it can still be a useful training tool in any point of the training process.

Training using Search (Discover) page describes how to use the search action sparingly. Avoid using it too much, as it can bias your model.

Search for terms or phrases in Explore in the same way as in Discover.

Key differences between using Search in Explore and Discover:

• In Explore you must review and label search results individually, rather than in bulk like Discover.
• Explore provides a helpful approximation of the number of messages that match your search terms. Check the example below for "cancellation" searching.

Search for a few relevant terms or phrases, and check how many approximate matches are in the dataset. Use this to estimate whether you have enough examples for a certain label.

Type your search term in the search box, at the top right of the page:

Example search query in Explore

To use Search in the Train tab, as another step from the Explore phase of the training:

1. Click the generic search recommendation in Train:
   
   

   

   
   
2. Select the label from the search list:
   
   

   

   
   
3. Review the LLM-powered label search suggestions.
4. Add the search term and preview the results (including the number of approximate matches):
   
   

   

   
   

Click to train these messages OR search for another term (before proceeding):

Note: The Batch has six results on a page, with typical Discover-like labelling experience (bulk + individual).

1. After labelling messages on the page, click Done.
   
   

   

   
   

A pop-up is displayed, with:

• a summary of training actions and options to close or
• search for examples for a different label, if the label doesn’t hit the criteria for no longer recommending search.