“We used to joke about Google helping you find your keys,” said Google’s vice president of search products and user experience, Marissa Mayer, during a Googleplex webcast this afternoon. Mayer was referring to the various advances Google is making in localized search products.
But the tremendous amount of geocentric data existing on the Web along with the subtle language and cultural nuances make organizing that information especially difficult. The inventor of the World Wide Web, Sir Tim Berners-Lee, has intimated similar troubles associated with the so-called semantic web because of its reliance on people to tag content consistently.
But generally, that doesn’t happen. “Perfect local search requires a near-perfect mirror world,” said Carter Maslan, a former Microsoft evangelist who relocated to Mountain View, Calf. Much of how a search engine would identify a place depends on how people have labeled it. The difficult part for robots is the various nuances.
New York, New York, for example is both a city in the New York and a casino in Las Vegas. This can make things difficult. “The first challenge is how do you identify a place?” said Maslan.
Google is tackling that very subject in several different areas. In image search, for example, the company is looking very closely at geolocation data associated with images. For Google News, the company has to decide if a news story is local news or national news about that location.
Google will look to story clusters, or bursts of stories popping up around the Web, to determine if the story should make the main page of Google News or should be rerouted to more localized areas.
Google will look to clustered data and clustered webpages to determine content for local search as well. Imagery comes from satellites, for example, and people use that imagery to create annotated maps.
“Annotating the planet,” though, is a big job and involves organizing information supplied by 350 million unique user activations and 50 million unique Maps/Earth users per month.
The future involves incorporating the 3D space as well, which would pull from terabytes of information annotated by people with local details and merging that information with satellite photos, aerial photos, and street imagery.
Politics makes it difficult, too. “When you start adding in disputed political boundaries,’ said Maslan, “you have to figure out where the person asking the question is in order to give a good answer.”
But precise (i.e., inerrant) local search that understands all the contexts of local information is still a ways off. Mastering this aspect of search means teaching a search engine the difference between a query for “Kansas State” and “Kansas,” the state, and allowing searchers to save zip codes so they can refine searches on their own.
David Utter contributed to this article.
