Approaching High Class Imbalance problem

I have a million documents, consists of text & other categorical fields and wanted to classify. Of these documents, I have only 30 class B observations and remaining are class A documents. Any suggestions on how you would approach? @jeremy