I second this, sounds like very good idea: Feedback for the algorithm so that it can update its internal model in supervised / reinforcement learning fashion.

I second this, sounds like very good idea: Feedback for the algorithm so that it can update its internal model in supervised / reinforcement learning fashion. Further ideas: users can suggest totally new tags for a given article, if the current tags seem to miss something.