I don't think it is necessary to identify the scraper. Something simple should work, like counting how many unique URLs for a particular website got accessed in

I don't think it is necessary to identify the scraper. Something simple should work, like counting how many unique URLs for a particular website got accessed in some period of time and comparing that to similar thing, but counted for known non-bot users (normalized). If one is much bigger than the other one - start requiring CAPTCHAs, but only for those requests, that are not part of known non-bot users set of our probabilistic counter of requests or even a separate bloom filter, depending on what is going to be more efficient and more accurate.