Does anyone have any advice on building an NLP model from a set of documents, such that you could ask “does this document discuss topic X”? I’d like synonyms and topics to be identified without supervision.