Recall measures the proportion of the total possible true positive results that the model was able to identify.
Recall = true positives
true positives + false negatives
For example, for every 100 verbatims which should have been labelled as ‘Request for information’, the recall would be the percentage that Re:infer successfully found.
A 77% recall would mean that for every 100 verbatims that should have had a specific label predicted, there would be 23 verbatims which should have been predicted as having the label, but Re:infer missed them.
For a more detailed explanation on how recall works, please see here.