Necessary TODO refactoring, improve errors system, etc. Ideas Try Marcov chains for article's text and title processing. Is it better CountVectorizer, or not? Use shelve insteadof pickle for serialization?