Seems pioneer was Reuters. The solution is based on using machine learning methods. First on the corresponding set of marked articles to build some classifier. He then used to assign new articles to one or another category or rubric that corresponds exactly to the problem of tagging.
Well, vskidku, just as an example:
Elaslic here is very far - only as a repository of information.
By the way, boasted to Reuters that he is on the implementation of this method saves millions, mostly on the payroll of the Department dispersed almost a hundred employees that they Tekirova news manually.