Google’s search team recently revealed to Wired’s Steven Levy some of its algorithm-related secrets. As many are well aware, Google’s algorithm is something it keeps very hush-hush about, so this was something of a rare occurrence.
The company’s search algorithm, explains Levy, in his multi-page feature article, is far from static. It is constantly being honed and changed and improved upon, specifically at Google’s weekly meetings at its headquarters, where dozens of engineers, product manager and executives discuss ways to make Google search better.
Levy lists some of the key advances in Google search since it first brought the ingenious Page Rank to the public. These include local connectivity analysis – brought about in 2003, giving more weight to links from authoritative sites; personalized results – started in 2005, users were given the choice whether to let Google mine their personal browsing behavior to provide customized results; and finally real-time search – which was brought in just last December, allowing users to receive search results from Twitter and blogs as they are published.
For the very few left in the world who are unfamiliar with Page Rank, this is what made Google famous: it is the ranking system invented by Google co-founder Larry Page that assigned a ranking to pages based on the number and quality of links to that page.
However, Google’s head of search Ubi Manber says that Page Rank isn’t the be-all and end-all of Google search. Google has skillfully learned how to incorporate contextual signals to help deliver the most relevant results. These contextual clues included the relevancy of the page title and the anchor text, as well as the freshness of the page, and the geographical location of where the content was created.
But some of the most important signals have been discovered democratically: Google gets millions of searches a day, generating data by clicking on certain results, changing their search query when they’re unsatisfied, and demonstrating the relationship between certain queries and certain geographical locations – these all provide Google with invaluable data that it then uses to improve its algorithm.
What Levy reveals is that with all this information being processed by the Google bots daily, the search engine is becoming very, very smart. It can understand semantics, like the fact that when water boils, it’s hot. It is also learning a vast amount of synonyms that are interchanged constantly in searches. It is even able to grasp the context within which a word is placed. This is the kind of intelligence humans may take for granted, but machines don’t have inherently. When they gain this kind of intelligence, they can become incredibly powerful (think of all the artificial intelligence movies out there).
The Google team is constantly on the look-out for problems with specific searches, and when they find them, they figure out how to fix them. The engineers test a tweak they’ve made on a small percentage of Google users, turning Google into something of a constant experimental lab. But that’s just how they roll. As the team continues to fix bug after bug, Google’s algorithm gets increasingly more intelligent.
The machine has even learned how to group certain words together (called bi-grams), depending on the context of the search. Levy gives the example of “New York,” which are two words that humans would instantly know should be grouped together, but a computer would not know this by default. Google does (now). It also knows that “New York Times” should be grouped as a three-word group, but that “New York Times Square” is a place, not a company. It’s a smart bot, that’s for sure.
As for the newly introduced real-time search, Google has, of course, discovered contextual signals for which Tweets should come out on top. The search engine is able to distinguish which Tweets are re-Tweeted, how many followers a Twitterer has, along with whether the Tweet is organic, or bot-generated.
If Levy’s article is anything to go by, we can expect our favorite search engine to continue to grow in intelligence, seemingly at exponential speeds.
Google isn’t matching words, explains Levy’s article, it’s matching meaning.
Kaila Krayewski is a freelance journalist with a passion for all things internet. Having worked for nearly two years as the public relations manager for an internation search engine optimization company, and publishing hundreds of articles (how-to, informational, and otherwise) on SEO, she knows a thing or two about the field. Furthermore, having just started up her own website blondetraveler.com, she is doing her best to keep one step ahead of the search engines in order to keep the traffic flowing.
Read other articles by Kaila Krayewski
Tags: google algorithm, google intelligence, google wired magazine
Very interesting. Now this Google Engine have some IQ and think on its own. I agree that its very good engine, but somehow don’t think that its so complicated. Still backlinks are most important and that means that engine from first version didn’t change allot.