As well as I see the code is still pretty specialised for texts classification. This is a good point for improvement.