Can I forge text data to train my model?

text_mining
classification
machine_learning
python
multi_label_class

#1

My model needs to classify short text data(less than 90 char) to 40 categories. however dataset is highly imbalanced(total 8000 rows, one of the categories has 1400 rows). I need to increase minority class amount but I do not want to purely do replication. Is there a better way?