I am working on a text classification problem. My data is highly imbalanced. For example, one category has 700 documents while the other has 30. I have around 30 categories. I tried different classifiers and the performance is consistently poor.

What is the best way to tackle this issue? Thanks