Hello. There is a task of developing a classifier classifier for speech style (artistic, journalistic, etc.). It is necessary to do this on the basis of the neural network. I can not determine the characteristics defining the style of the text and figure out how to translate the source text in the characteristics vector with which the classifier will eventually work.

How can this be done?

Answer 1, Authority 100%

Here Table:
(Picture Tables taken hence )

As can be seen from it, in the texts of different styles, either terms or descriptions or appeals, or slang or other classes of words can prevail. Classes of words will determine the neural network after training. Styles it will also determine, recognizing the predominance of a neboral class or a combination of classes of words in the text recognition. The text structure will also be among the signs (at least at the level structured / not structured).
And then – literature about learning neural networks with reinforcement and training with the teacher.

Most importantly – Algorithm for reverse error . (https://habrahabr.ru/post/154369/ – on Habré).