ETech: Mobile Phones Reveal the Behaviors of Places and People

[Quinn managed to scoop me blogging about Tony Jebara’s presentation! But after I chatted with her, we both agreed that I should continue with my blog post and see if I can augment her post a little.]

Tony Jebara’s presentation “Mobile Phones Reveal the Behaviors of Places and People” really opened my eyes to what amazing things you can derive from large data sets. Tony co-founded Sense Networks which specializes in taking GPS and mobile phone location data and deriving as much useful information as possible from it. Sense Networks works with mobile phone service providers who collected data from users who opted in to have their data be collected and mined. All the data they receive from the service provider is GPS location data — no personal information at all was ever made available to Sense Networks.

Using mobile location data passively, the goal is to build a more realistic Facebook. More realistic in the sense that it should mirror real life connections and not the 2000 random friend connections you might have on Facebook. Based on mobile phones co-locating, Sense Networks hopes to deduce a real life friendship and then enable intelligent online applications.

For instance, Sense Networks can can deduce the activity of bankers in the financial district of San Francisco. Given that a cell phone moved to the financial district in the morning and stayed there for 8 hours, they assumed the signal came from a banker. Associating signals to movement patterns allow Sense Networks to start correlating the DOW and when bankers arrive at work. Apparently when the DOW is up, bankers are more likely to head into work sooner, but when the DOW is down, bankers are likely to take their time getting to work. Tony’s team can also correlate between night life and the health of the DOW — when the DOW is healthy more people move from work to popular night time spots, and less so when the DOW is down.

The next thing that Sense Networks wanted to figure out is where similar people to you might be. Tony suggests that if you build a network of people based on co-location data and examine the location at which people congregate you can start to group people into like minded tribes of people. To do this, Tony’s team needed to build a network of places that determines the flow of people in and flow of people out of a place. Commerce information of what types of business are located in a given location and demographics for a location give away a surprising amount of information about the people who frequent those places. This information then allows Sense Networks to determine what is happening on any given street corner.

Putting this data into a matrix of hours in a week and probabilities of a user engaging in the various activities yields many interesting results. For instance, you can deduce a lot from a person simply by looking at where the person sleeps at night. Do they sleep in a wealthy, single or family neighborhood? Or, if a person hangs out in wealthy neighborhoods, but lives in a poor neighborhood you can deduce a certain aspirations of the user. This type of analysis yields a very general model of people and Tony emphasizes that the personal data on any given user is discarded and that only general trends of behavior are sought.

I find it fascinating that these kinds of trends can be deduced from GPS, commerce and demographic data. While I find the technology and data clustering algorithms that enable these features very interesting, I have to question how the results will be used. Tony assured the audience that their customers will use the data to better target advertising, But I already have enough pesky advertising I’m working hard to ignore — I’m not certain I need more targeted ads bombarded at me. But there are loads of scary applications for this data as well — the number of Big Brother type applications enabled by this scare me.

Regardless of how this turns out, I really appreciated Tony’s presentation because it enlightened me to some of the possibilities of mining large datasets for valuable nuggets of information.

tags: , , ,