SAN FRANCISCO — Google on Wednesday began making its search engine smarter in a major upgrade that looks beyond query words to figure out what people are actually seeking online.
“Knowledge Graph” technology built to recognize people, places or things signified by keywords took its fledgling steps in the United States with the hope of eventually extending it to Google searches worldwide.
“The Knowledge Graph is built to understand real things in the world,” said Google fellow Ben Gomes, who has worked on search at the California-based company for more than a decade.
“It is the beginning of a long journey we will be on to cover more topics and more complex queries.”
Gomes envisions Google search being able to eventually answer tricky questions such as where to attend an outdoor Lady Gaga concert in warm weather or the location of an amusement park near a vegetarian restaurant.
For now, people using US Google search in English will start seeing on search pages boxes suggesting what they are interested in finding.
A demonstration showed that searching on the word “Kings” in California, for example, prompted the search engine to point out that one is likely interested in a hockey team, basketball team, or film.
Using the keyword “Andromeda” prompted Knowledge Graph to note one might be interested in a galaxy, a television series, or a Swedish rock band.
Clicking on a suggested topic instantly refined search results.
Google painstakingly adjusted its algorithm to comb information from databases such as Freebase and Wikipedia to give context to words and then use general search patterns when it comes to what people tend to want, Gomes said.
Searches on specific subjects such as an architect’s name triggered Knowledge Graph results offering to dive into categories such as biographical information or projects designed.
Google added a serendipity factor by surfacing potentially surprising facts.
For example, a search on “Simpsons” cartoon creator Matt Groening resulted in a Knowledge Graph box that noted his parents and sister have the same first names as his well-know fictional characters — Homer, Margaret and Lisa.
“Text strings are ambiguous; we have to make a lot of changes to understand real world entities,” Gomes said.
“This has been an exciting problem for us and we have been attacking it for two years.”
Google’s Knowledge Graph has been programmed to recognize more than 500 people, places, or things using a combined total of about 3.5 billion attributes and associations between bits of information.
“People ask about anything you can think about and lots of things you never thought about,” Gomes said.
Google included links searchers can click to point out when the Knowledge Graph gets something wrong.
“Not everything is going to be correct,” Gomes said. “We put a tremendous amount of work into it but even if it was perfect facts change every day.”
The change was expected to affect a large number of queries, and was tailored with mobile gadgets in mind since it lets people dive deeper into searches with taps of touchscreens.
“It is very useful on a mobile phone, and really cool on a tablet,” Gomes said.