User permissions required: 'View Sources' AND 'Review and label'

 

If you have a pre-existing taxonomy, with many reviewed verbatims, adding a new label requires some additional training to bring it in line with the rest of the labels in the taxonomy.

 

When adding a new label to a well-trained taxonomy, you need to make sure to apply it to previously reviewed verbatims if the label is relevant to them. 

 

If you do not, the model will have effectively been taught that the new label should not apply to them and will struggle to predict the new label confidently.

 

To do this:

  • Create the new label
  • Use the search function in Explore or Discover to find similar instances. Think of key words of phrases within the verbatims that would be associated with this label to find more examples
  • Once you have labelled quite a few examples using search and some time has elapsed allowing the model to retrain, use the Teach functionality in Explore on that label and filter to 'reviewed' verbatims
  • This will show you reviewed verbatims where the model is unsure why this label does or does not apply
  • You should look for examples where the model thinks that the new label does apply, but has not yet been applied
  • Apply the label to all of the verbatims that the model correctly thinks the label should have been applied to
  • Keep training in this page until you have labelled all of the correct examples, and Teach no longer shows you examples where the label should actually apply

 

You can check how the new label performs in the Validation page (once the model has had time to retrain and calculate the new validation statistics) and see if more training is required.

 


Previous: Reorganising your Taxonomy     |      Next: Maintaining a Model in production