How to weed out irrelevant texts?

Question

Algorithms

Good afternoon.
There is a set of texts. This set contains insignificant messages from the category "+100500", "I am of the same opinion", as well as significant ones "Angara cost the state too much", "What happened in Siberia is simply amazing".
Those. significance, for a given task, is determined by the ability to "attach" a message to a specific topic. It is required roughly (quickly) to select significant texts.
Tell me, what approaches can be?
So far, it occurred to me only with the help of mystem/phpmorphy to determine which parts of speech in what ratio are found and by this coefficient. filter... However, this will obviously not be very efficient and may discard meaningful texts...