I am using WEKA for the text classification task. I have data (few thousands articles) to classify in positive or negative class.
In learning dataset I have 200 articles (12 positive and 188 negative) and with this ratio the result is not good.
My question is:
"What ratio of positive and negative articles in learning dataset will be perfect for the accuracy?"
Any rule, suggestion etc.

Thanks in anticipation!
Regards/
Sardar