Ask HN: Unsupervised text classification?

Hi everyone, We are currently using text classification model trained on a labelled data set consisting of almost 500 hierarchical categories , For example a news article is labelled as News/Business News , now what we want to do is extract sub-topics from this main category such as Stock Market , Trading News, Business Investment, Angel Investments etc . So what approach would be recommended to handle this kind of problem ?

Thanks