The public is relatively comfortable sharing information across the social sphere, barring spying by government entities, and companies are finding new ways to figure out what people want through their status updates and tweets. At last week’s Glimpse NYC social discovery conference, a panel of executives from EasilyDo, Yelp, Tagged, and Microsoft, moderated by Cadie Thompson of CNBC, looked at how these companies use big and little data to improve the recommendations offered through their apps and services.
A daily deluge of data comes Microsoft’s way, said Stefan Weitz, senior director with Bing, in the form of 2 billion Facebook status updates, half a billion tweets, 5 million check-ins from Foursquare, and other social info from the likes of Klout, Quora, and Google+.
Before privacy watchdogs could cry foul, Weitz said in terms of Facebook data, Bing has access to anything that has been authorized by users and is not explicitly private. He also said users should be mindful about information privacy—with realistic expectations. “Everybody does a good job of taking care of their users’ data, some are better at it than others, but nobody can do it perfectly,” he said.
Of the information that Bing can index, he said there is plenty of useful implicit data associated with the content people make available socially. “If you tagged the location from where you posted your update, posted a photo with people’s names in it, if you even have words in a photo we can OCR [optical character recognition] that and figure out what that is,” he said.
Mining social information to offer recommendations is not merely about knowing that someone posted from a Shake Shack, he said. “It’s the time of day, who you’re with, which Shake Shack you’re at,” Weitz said. “All those variables come into play.”
Such data gets put to work with the Windows Phone platform, for example, through a recommendation function called Local Scout. The software looks at where the user is and then runs an algorithm to check what their friends are up to, where their friends have checked in, and what they have liked. Weitz says Lcoal Scout also checks to see what events and places are popular in the city based on tweets and likes from the anonymous masses.
Local Scout compares that information with the user’s search history as well as prior check-ins and likes to offer up places and events that may be of interest. “In about 35 milliseconds, we’ve pulled together a list that says, ‘Here are 20 things we think you’re going to like within a two-mile radius of where you currently are,’” he said.
While recommendations are not new, Weitz said Microsoft is using its technology to show users why they may like the suggestions. “The problem with social recommendations is that in many cases you don’t why you’re seeing them,” he said. By explaining that a restaurant was suggested based on the interests of a particular friend, Weitz said the user can decide if that friend’s tastes actually match their own. If users dismiss recommendations in Windows Phone, he said, those choices get marked in their profiles to improve future results.
Data from social discovery may be overflowing but Mikael Berner, CEO of EasilyDo in Mountain View, CA, specializes in the minute details. He believes there is an untapped opportunity in using small, specific data sifted from heaps of information. “You tend to only see about 12 percent of what is in your Facebook feeds,” he said. “We try to find the 12 percent you care about, such as the friend getting married.”
EasilyDo developed a smart assistant mobile app that helps users get organized and save time. While big data can help develop certain recommendations, Berner said small data can pinpoint the direct needs of users. “Knowing that someone has to be in a hotel within three miles of an area tomorrow night, you don’t have to use big data to try and predict that,” he said.
Refining the information collected from the growing number of sources can improve recommendations to users, said Eric Singley, vice president of consumer and mobile products with San Francisco-based Yelp. “It’s tricky to get that balance right,” he said, “but if you have a lot of data available you can do new things.”
The reviews of local business posted on Yelp may show comments from friends up top, Singley said, but there is also a broader consensus of reviews from the overall community. “The practical reality is your friends probably haven’t reviewed every plumber in New York City,” he said.
Singley believes there is more to gain by marrying large amounts of data to highly specific information; in June, Yelp relaunched the Nearby function of its app that shows recommendations of places close by the user. For example, if it is raining, basic weather data can be combined by the revamped app with info on the best movie theaters in the area for recommendations, he said.
As more people share information on social sites the discovery process will improve, said Johann Schleier-Smith, CTO and co-founder of San Francisco’s Tagged. His site helps users find new friends and share their interests. “We build out profiles of who people are, not just based on what they say but how they interact with others,” Schleier-Smith said. Tagged has an algorithm that can make recommendations based on those interactions without knowing users’ age, gender, or sexual orientation, he said.
Schleier-Smith is a fan of using algorithms and computing power to create useful social data and said there are ways to discern what people favor even through dating apps such as Tinder. “How long it takes to swipe [yes or no] means how carefully you are looking,” he said. “What is your degree of interest? We can find a lot of patterns in that data.”
