How Google’s RankBrain uses artificial intelligence to handle 15% of search queries
Google has announced that it is using artificial intelligence (AI) to help it handle up to 15 per cent of search queries.
The part of the company’s search algorithm, which has been nicknamed “RankBrain” internally, is thought to be Google’s largest deployment of AI to date.
Google is the world’s most popular search engine. It receives more than 3 billion search queries per day globally. In this part of the world it enjoys a virtual monopoly, with a market share of up to 97 per cent.
AI and machine learning have played an important role in search for some time. But Google’s main search algorithm is thought to still rely heavily on human-generated code, which of course has its limitations, especially when dealing with never-seen-before search queries.
Speaking to the Bloomberg news agency earlier this week, Greg Corrado, a Google research scientist specialising in AI and machine learning, said RankBrain had been interpreting a “large fraction” of queries for the past few months.
Google uses 200 or so signals to determine where pages should rank for a given query. While RankBrain is just one of these signals, Corrodo revealed that it had become the third most important since its rollout.
“I would describe this as having gone better than we would have expected,” he admitted.
The use of AI and machine learning feels like a natural next step after the release of Google’s Hummingbird algorithm.
When it first went live back in 2013, Google said Hummingbird would interrogate queries and return pages not just based on particular keywords, but on the assumed meaning behind full search terms.
This was nothing new. Google had been using its vast index to make assumptions about what users want since the launch of universal search, when it started to serve up different types of results (images, videos, maps etc) without needing the user to specifically ask for them.
But that was all based on Google’s Knowledge Graph: this piece of information relates to that piece of information, so we can make your search results richer.
What’s different about RankBrain is that it’s not using existing “knowledge” held by Google. It’s an additional step in the process, giving machines, rather than human-generated code, more of a say in what users see when they run a search Google hasn’t seen before.
Maintaining the edge in search
Keeping its search offering ahead of the competition is absolutely vital to Google. That fact was further underlined earlier this month when the company announced its financial results for the third quarter.
Google made USD $16.8 billion from advertising in the three months to the end of September. That represented 90 per cent of its total revenue.
Continuing to hit those sorts of numbers requires Google to remain the go-to when you want to find something online. Constantly adjusting its search algorithm to more accurately return what users are looking for is therefore a top priority.
As well as rising expectations among users and an ever-expanding internet to crawl and index, Google faces challenges from rival search providers.
Facebook, for example, has recently revamped its search offering. The world’s largest social network is keen to leverage the two trillion posts it has indexed from its users to claim an even larger share of the total time people spend online.
Many of these posts are behind a Facebook log-in and cannot therefore be indexed by Google or other search engines.
As well as access to this private treasure trove of content, Facebook also has a lot more information about each user. As well as the search query they’ve used, Facebook will also have personal details, from who their friends are to places they’ve visited, to help shape its search results.