Of course, while RankBrain is primarily used by Google on the 15% of new-to-Google search queries, don't be fooled into thinking that is all it does. Because it is (to quote Google), "Not limited to any particular set of queries". In other words, the technology driving RankBrain also comes into play when trying to make sense of other long-tail searches. And post Hummingbird, long-tail Google searches make up over 40% (and growing) of all search queries. And as keyword driven search is dying (and it is) and question based and vocal search becomes more and more the norm, long-tail question driven search queries are more and more where the SEO gold is buried.

Google's Gary Illyes was asked on Twitter to clarify whether the RankBrain algorithm continually re-programs itself and learns as it goes along (as was originally believed by the SEO community). But it turns out it doesn't. Rather, RankBrain is periodically 'Re-trained" by  the boffins at Google, to do its job better, as and when new up-dates are released. In Illyes own words: "Its effects are expectable, not assumable".

IIlyes went on to say: “We’ll keep experimenting with and testing new models, and we’ll make updates as we come up with models that do a better job than the existing one. That could be about refreshing the data or developing new neural net architectures.”

Another ingredient in RankBrain is believed to be Google’s Word2Vec Project, which is based on the twin architectural models of CBOW (Continuous Bag of Words) and Skip-Gram. Both of which are, in nerd-speak, considered “shallow neural models”. In essence, CBOW tries its level best to predict the current word, based on the words around it (because words make more send when viewed in context with the words nearby). Whereas Skip-Gram cleverly predicts meaning from the sense of the words it finds.

Or to simplify it even further:
​

• CBOW predicts the word given its context.
• Skip-Gram predicts the context given a word.

Whether RankBrain is actually using Word2vec (or a variation of the methodology) or not, is ultimately by-the-by. All we really need to know is that RankBrain is converting words and phrases into vectors, which it then uses to allow for a deeper understanding of the data it's reviewing. The implications of all this being that if Google is able to convert a typed question into vectors to better understand the inherent meaning behind the question, then Google will be able to significantly improve the search results it shows. 

Another potential element under-pinning RankBrain (and until Google opens their Kimono to reveal all, this is just more guess work on our part, remember) is thought to be ‘Distributional Hypothesis’.

In linguistics, Distributional Hypothesis states that words used together in a sentence tend to infer similar meanings. This is a theory born from Statistical Semantics. The distributional hypothesis for this infers that the more semantically similar two words are, the more distributionally alike they’ll be, and the higher probability that they occupy similar linguistic contexts.

Or to simplify it even further:
​

• Distributional Hypothesis states that a word is more often than not characterised by the company it keeps.

Ultimately RankBrain will expand its remit from learning patterns within short strings of data, to learning patterns within entire documents. Because just as there is learning to be had in the former, there is surely deep learning to be had with the latter. And it's not too far of a stretch to think that the RankBrain algorithm of a few years down the track will be able to read an entire book and distill its meaning into a few short paragraphs.

Cliff Notes look out!

Okay, so let's back track for a minute. RankBrain “Converts words and phrases into vectors," right? But what exactly does that mean? To better understand that, we need to look to Professor Geoff Hinton of the University of Toronto's work in 'Word2vec Connections'. Mr. Hinton is a gentleman and scholar (as the old saying goes) with much wisdom in the field of artificial neural networks. And who happened to have owned a company called DNNresearch Inc, which was acquired by Google in June 2013.

(Ah...the plot thickens...)

Neural Net reporter, Jack Clark of Bloomberg (who wrote the original Bloomberg Business article) attempted to get clarification on the whole Word2ve Connection issue:

"They wouldn’t explicitly confirm that it (RankBrain) is Word2vec, but everything we discussed indicated it’s likely doing something roughly equivalent to Word2vec, and is also doing similar conversions for sequences which is likely connected to Sequence to Sequence learning (PDF: http://papers.nips.cc/paper/5346-sequence-to-sequence-learni…).

It also links to Geoff Hinton’s stuff on Thought Vectors which implicitly involves word2vec."

When pushed on the subject, a Google spokesperson said “It’s related to Word2vec in that it uses ’embeddings’ — looking at phrases in high-dimensional space to learn how they’re related to one another.” Which is about as much clarification on the subject as we could rightfully expect to get.
​
​For those with a mind to dig deeper into Rankbrain, there's a Patent Application by Googe pending, which includes this:

Which segues us rather nicely back to the question of, 'Okay, so if RankBrain is the third most important factor in Google search, what are the first two?'.

This we didn't need last week's announcement to answer; because I, and many other SEO types, have known the answer to this question for years. The first most important thing (in the post Panda world) is QUALITY ORIGINAL CONTENT.  The second most important thing (in the post Penguin world) is QUALITY LINKS.

You'll note I added the word 'Quality' to both those Google ranking factors? Good. Because if you think all you have to do to rank in Google is flood your website with cheap content (no doubt sourced out of Eastern Europe), and point thousands of low value links at it (no doubt sourced via an el-cheapo company in India), then you're probably the proud owner of a time machine*, and are kickin' it back in 2010 listening to Ke$ha's mega-hit, 'TiK ToK', or else queuing up with the kids to see Toy Story 3 at the cinema for the umpteenth time. I say this as nobody with half a brain in their head would run that strategy today. Because if they did, their website would have the living crap kicked out of it by both Panda and Penguin, and find itself buried on page 35 of their particular Google search. And given that 92% of people don't go past the first page of Google...good luck with getting the phone to ring from there.

The most advanced research vessel in British history is on the brink of being Christened RSS Boaty McBoatface. Yes, you read that correctly. Boaty McBoatface! The boffins over at the National Environment Research Council thought they'd be a bunch of clever clogs and run an on-line poll to name their new multi-million dollar research vessel. I mean, what could possibly go wrong?

Well, you know...being the internet, everything that could go wrong probably will. To paraphrase Voltaire: "The trouble with common sense is, not many people have it'. Which is another way of saying, never trust anything you read on the internet!*

*unless you read it here of course...then naturally the information must be gospel! ;)

And if Boaty McBoatface wasn't enough of an incentive to stop people running on-line polls to name things, now the US Air Force have joined in...

So expect 'Planey McPlaney' to put the fear of God into those terrorists over in the middle east sometime soon!

Big thank you to Belinda and all the amazing staff at Fancy That Costumes in Crows Nest. Super helpful staff! Couldn't have got the costume for our new corporate branding, 'Logan Thor - The Superhero of Search' done without them!