How to detect novelty in textual data streams? A comparative study of existing methods