06 June 2012
SEARCH engines have barely changed since Google was founded in 1998. Sure, they run on blazingly fast servers and are powered by sophisticated algorithms, but the experience itself is basically the same: users enter a word or two and the engine spits out links to the most relevant pages.
That is about to change. Last month, Google rolled out its “knowledge graph”, which serves up facts and services in response to search terms – not just links to websites. It is the latest step in a process in which search engines are morphing into something quite new: vast brains that respond directly to questions posed in everyday language.
“Search does a good job of returning pages,” says Shashidhar Thakur of Google. “But we can go beyond that and return knowledge.”
Links are not necessarily the best way to answer a query. When I search for “location of Arsenal Football Club”, for example, I would prefer to get a direct answer telling me the address of the club’s ground in London, rather than a link to a document containing the information. Google and Microsoft’s Bing can already provide direct answers to a small number of queries, but the range and depth of those answers is about to expand dramatically.
Over the past few years, Google and Microsoft have been quietly compiling vast knowledge databases to help them do this. Their stores have been built from publicly available information, such as Wikipedia pages, as well as prices from retail websites and user reviews. “All the things that describe an object are scattered across the web,” says Stefan Weitz, director of Bing Search. “The challenge is to bring them together.”
This open-source data is combined with internally acquired data, such as location information from the firms’ respective mapping products. It is all put together to create a “graph”: a network of things, such as people and places, and the relationships between them. Google’s graph contains 500 million entities linked by tens of thousands of different types of relationships, Thakur says. Microsoft’s knowledge graph, which it calls the Satori database, contains 350 million entities, says Weitz.
At Google, algorithms now trawl the graph for information as well as searching the web. When I enter “beagle”, for example, I am shown a panel of data on the dog breed as well as links to relevant websites. The selection of data is based on previous searches on this topic, so Google knows in this case that users are interested in things like average height and lifespan.
Microsoft is taking a different tack. Its Snapshot service, due to be added to Bing this month, will use its knowledge graph to serve up links to services associated with the search term. A search for a restaurant, for example, will still return the restaurant’s website, but when a user moves the mouse over the link, Bing will give them the option of reading reviews, seeing a picture of the venue or booking a table there. Weitz says that the aim is to guess the real-world action that a user is interested in when they search and to return links that allow them to carry out those actions.
Both Google and Microsoft are also expanding the ability of their search engines to understand queries phrased in everyday language. Combined with the knowledge graph, new kinds of search will become possible. Thakur is working to make Google’s search engine answer detailed questions about the world, such as listing all books that have been made into movies, or all volcanoes that erupted between specific dates.
Microsoft’s service-oriented approach will be particularly suited to mobile devices, says Weitz. People on the move often want a direct answer to a question, like “how do I get to the nearest subway station?”, rather than a link to a page with a relevant map.
The two giants face competition from a recent arrival that has formed a powerful partnership with one of their rivals. Wolfram Alpha, the creation of mathematician and entrepreneur Stephen Wolfram, also relies on a vast knowledge store. But unlike at Google and Microsoft, where data is often pulled from websites, the 200-strong Alpha team assembles its knowledge graph from traditional, fact-checked sources and eschews use of Wikipedia. Alpha can also perform data mash-ups, such as comparing stock prices. By drawing on geolocation and air travel data it can even identify the planes flying overhead.
Alpha got a break last year when Apple announced that it was using it as the brains behind Siri, an automated assistant that comes with the latest iPhone. It is the kind of application that Google and Microsoft, which have their own voice-recognition systems, are also looking at.
And Siri may be just the beginning. Voice-activated devices could soon tap into knowledge graphs to do all sorts of things. Want to listen to that 1980s pop song that was on the soundtrack of that action flick you have forgotten the name of? Or identify the digital camera that costs less than $150 and has the best user reviews? You have to trawl a list of links to get an answer, for now, but it will not be long before the graph delivers one in an instant.