Etech liveblogging: Mobile Phones Reveal the Behaviors of Places and People (Tony Jebara)

(Tony is from Sense Networks, and also a prof at Columbia University in comp sci)

Starting out with what we have now:

Online data isn’t disconnected documents, but a network, with links between docs and the key information is the links. Folks like Google have obviously exploited that network technology. Online social networks, networks of people, the relationships being the important part. – looking at Facebook, but also affinity networks like Amazon recommendations.

The issue is using some real world activity to build networks. What can you tell about a place by what it’s connected to? They’re using mobile location data passively – but it’s messy and hard by comparison to online data. Should Facebook be able to build my friend network by seeing our phones cluster? (That will put a damper on my extramarital affairs.)

We already have smarts in online data: collaborative filtering, marketing, advertising, search, social recommendation – the next step is pulling that out of location data.

They’ve been collecting location data from taxis, blackberries and iphones

An example of what can be found:

In SF – commuting into the financial district for work. When people come into work correlated with the Dow Jones, when the stock market started going down people rolled into work later.

How much are people in SF going out at night? How late are people staying out? Night life goes way down with the economic downturn, in fact gps is down over 30% in general in San Francisco now.

The app: citysense.com: Where is everyone? How is sf right now? They can show you a heat map of iphone or blackberry to get a feeling how active the world out there is-
this brings the ability one has in a small community to tell if something is going on to a scaled up urban area. Go towards the red dots!

You can search for bars and restaurants in ranked order in how active they are. They have a buddy finder: kind of like a Google latitude.

In the next step, citysense 2.0 they are color coding the dots to find people like you. Each color represents a different ‘tribe’ – orange is the young edgies, light blue the business traveller, for instance. The citysense app determines what crowd you’re in – this is the ‘secret sauce’ – tehy’ree going to try and build a social network out of the location data. It’s honest in a way Facebook isn’t, because you co-locate with your friends. Both actual colocation and behavioral colocation- if you go to the same kind of place at the same kind of time, that’s a semantic relaitonship.

They start by building a network of places- like google meta data but for physical locations. For every possible place or street corner they’re looking to find is place a similar to place b? Some of this can be got from gov databases, and some from flow analysis. Similar to page rank… if people come to a place from similar types of origins, then leave to similar places later, they can extract that as meta data about the place. eg bankers leave the Financial District, go to an Italian restaurant then go to a similar neighborhood for the night.

They color code bars by the similar inflow and outflow, so they make them semantically adjacent. They are working with an advertising company to change how they target their beer ads.

Then to the poeple: they translate the gps trails of users into flows. As in, what are the odds of finding person of demographic 2 in commercial sector 3, which is fine dining, at 6pm on saturdays: 52%

Measure them not where they live, but where they hang out on average as a probability. They then toss out the actual location data and only keep the matrix. The matrix light up quickly because we all follow very normal weekly routines. (The Dopplr crowd must look super strange to them)

They have 4 million users – with semantic data relationships being the social network.

They identify tribes based on advertising applications:
young and edgy: poor, more ethnically diverse
weekend mole- out occasionally on weekdays, Latino, middle income neighborhood
mature homebody – rarely goes out

This is to help companies better target their ads.

They’ve had to do interesting re-calibrations recently. Usually the season requires re-calibration, but the economy has caused massive changes.

They’re interested in the next network: it’s not the online network, it’s the offline world.

Comments are closed.