rebalancing of the categories in the training data

[strayMat]

The method that I heared of belongs to the category of 'sampling methods that manipulate the training data to change the class distribution.' A successful example is the SMOTE method (Chawla et al., 2002). The general idea is to artificially generate new examples of the minority class using the nearest neighbors of these cases. Furthermore, the majority class examples are also under-sampled, leading to a more balanced dataset.

This is a data augmentation method that we could make a good deal of to augment the size of our dataset.
PB: It is not trivial to learn to generate plausible sentences from under-sampled categories.