Martin Splitt from Google said Google may be using machine learning in crawling in production right now, or it might be just an experiment, but here is how it does use machine learning for crawling in search. I should note, as far as he is aware, machine learning is not used for rendering in Google Search.
How does Google uses machine learning for crawling? Two ways, for predicting quality of the URL before Google crawls it and for predicting the freshness of the URL before Google crawls it.
“I know we are using machine learning to identify or to predict what we will get from a crawl in terms of quality,” Martin Splitt of Google said. He said it is interesting to try to “predict what kind of quality we can get from a specific crawl before it happens.” He said the reason for that is because it would allow Google to “schedule crawling more intelligently.”
Should Google bother using resources to crawl a URL that might not be worth its time, because the page is not high enough quality to warrant it is the question the ML is trying to answer.
Martin Splitt said “the same goes for if we can predict freshness, as in do we know that we have to schedule this web site daily or is there, can we gather signals to tell us not to do this daily.”
So Google can use machine learning to predict this for sites where Google does not have that much data for, i.e. content change frequency data.
He clarified “but I am not sure if we are using this in production or this is an experimental so far.”
Here is the video embed at the start time:
Forum discussion at YouTube Community.