Machine learning LDA, text classification

I need to classify texts (news) by the importance of the Russian Federation. 1 category-threat of NB (key phrases: terrorism, nuclear weapons, etc.) 2 category-CIS countries (key words of the name of the countries ) 3 category of NATO G7, etc. Using the standard method, you send a set of texts, create an LDA model, and determine the number of topics to divide these texts into. And LDA creates a list of topics to which it assigns individual words from those texts with the weight of importance!. I need to do it myself with your own hands, create a list of words with the weight of importance and then use it to train the program, and not vice versa. Who knows how to do this, please tell me, or if anyone knows another way.

Here in the picture in the bottom line is the list that I need to fill out myself. But I have little experience, and I can't figure out how to fill it out myself, without delivering a set of texts