Train ibm watson discovery

Question

I have 459 json documents in my discovery collection. I want to train discovery by making some exemples of question and apply ratings to my results: "Relevant" and "Not relevant".
I woulk like to know if I must rate all documents as Relevant or Not relevant results for a query (max 100 doc) or I can rate the Relevent document and some other (10, 15, ...) documents as Not relevant .
Thanks

Answer 1

Currently you must label at least 49 positive examples as well as "a few" negative examples. Any combination of relevant and not relevant ratings across all your queries should be fine. There is some documentation available, which describes specific training requirement in detail: https://console.bluemix.net/docs/services/discovery/train-tips.html#relevancy-tips