Pages

Tuesday, November 24, 2015

RankBrain: How Google Is Using Artificial Intelligence To Rank Web Pages




The days of a hand-crafted human search algorithm may soon be over. Google has announced RankBrain, a new artificial intelligence that is now processing a “very large fraction” of searches each day. This new system was reported to Bloomberg and has been implemented, in part, to help Google better understand deal with the amount of ambiguous queries that they receive each day.

RankBrain isn’t replacing the current Google Algorithm (currently nicknamed Hummingbird), but instead RankBrain is one of the “hundreds” of signals according to Bloomberg. Previously Google has claimed up to 10,000 signals & sub-signals. It turns out that the RankBrain signal is pretty important too. Greg Corrado, Google senior research scientist, stated that RankBrain is now the 3rd most important signal in matching results to a search query. When asked about the other two Google declined to comment.

The artificial intelligence of RankBrain comes in the form of mathematical entities called vectors that can be understood by computers. When presented with an unfamiliar word, RankBrain will help formulate a guess at what the query was about and filter accordingly. At Search Engine Land we took a deep dive into RankBrain and were given the following Google authored post on machine learning with some examples for those looking to learn more on the topic.

So now you may be asking yourself, When will I see this RankBrain in my results? The answer is that you most likely have already. The given example of a search query that leverages RankBrain is “What’s the title of the consumer at the highest level of a food chain”.

On Search Engine Land it was pointed out that the lengthy queries may be simplifying the search and making more manageable.

At first glance, RankBrain doesn’t appear as much as a signal as it does a refinement. Google confirmed that RankBrain is a signal, but it is also important to understand the complexity that a signal can be. Things like links or words on a page can be signals and signals can be associated with users as well. There are many possibilities as to how RankBrain could work into being a signal, especially if it takes a variety of additional elements into account to help classify the right content for users.