Google team proposes way to rank search results by their accuracy

Company's computer scientists devising ways to rank search results not by hits, but by accuracy

PUBLISHED : Wednesday, 04 March, 2015, 7:43am
UPDATED : Thursday, 16 April, 2015, 12:10pm

The internet, we know all too well, is a cesspool of rumour and chicanery.

But in a research paper published by Google in February - and reported over the weekend by New Scientist - that could, at least hypothetically, change. A team of computer scientists at Google has proposed a way to rank search results not by how popular web pages are, but by their factual accuracy.

To be really clear, this is 100 per cent theoretical: It's a research paper, not a product announcement. Still, the possibility that a search engine could effectively evaluate truth, traditionally an exclusively human domain, promises a fundamental change.

It's not too difficult for computers to determine whether a given statement is true or false. To evaluate a stated fact, you only need two things: the fact and a reference work to compare it to. Google already has the beginnings of that reference work, in the form of its Knowledge Graph - the thing that displays "August 15, 1990" when you search "Jennifer Lawrence birthday", or "American" when you search "Obama nationality".

Google culls those details largely from services such as Freebase, Wikipedia and the CIA World Factbook; a separate, internal research database, called Knowledge Vault, can also automatically extract facts from the text on web pages.

Whichever database is used, Google structures these facts as "knowledge triples": subject, relationship, attribute. For instance: Jennifer Lawrence, birthday, August 15 1990.

To check if a fact is accurate, all Google has to do is reference it against the knowledge triples in its giant internal database, and to check whether a website is accurate, Google would just look at all the site's knowledge triples and see how many don't agree with its established body of facts.

The distant suggestion, these researchers write, is that Google's version of the truth would iterate over time. At some point, perhaps even Google's hotly debated and much-studied ranking algorithm could begin including accuracy among the factors it uses to choose the search results you see.

That could be a major event: In one trial with a random sampling of pages, researchers found that only 20 of 85 factually correct sites were ranked highly under Google's current scheme. A switch could, theoretically, put better and more reliable information in the path of the millions of people who use Google.

In that regard, it could have implications not only for search engine optimisation - but for civil society and media literacy.

It's worth noting that the Barack Obama nationality example comes straight from the Google report, which would seem to imply that the technology's creators envision it as a tool against stubborn misconceptions and conspiracy theories.

"How do you correct people's misconceptions?" Matt Stempeck, the guy behind LazyTruth, asked New Scientist recently. "People get very defensive. (But) if they're searching for the answer on Google they might be in a much more receptive state."

Just three weeks ago, Google began displaying physician-vetted health information directly in search results, consulting with the Mayo Clinic "for accuracy".

It's unclear exactly what Google plans to do with this new technology, if anything at all. But it would certainly give new meaning to the phrase "let me Google that for you".