Mobile 2.0 Europe: Context Panel

Moderator: Raimo van der Klein, SPRXMobile

Gregr Skibiski, Sense Networks

Tommy Ahers, Vodafone (ex-Zyb)

Felix Petersen, Nokia Berlin

Ted Morgan, SkyHook

Xavier Carrillo Costa, Digital Legends

XCC: we spend a lot of money simulating the real world. Now we're connecting with the real world, and it's changing the game paradigm.

TM: Have difficulty figuring out some behaviour. We see some locations which have huge numbers of lookups, but can't work out why. One of our heaviest location users on the iPhone is the RIM HQ.

FP: We have a successful mapping product. We realised a couple of years back that we didn't need to license traffic data, we could gather it ourselves from looking at behaviour of users with our product. Google built a product on the back of inferring meaning from links between pages; we can do the same with geographic links.

TA: Heading up location team at Vodafone now. When I promised to change Vodafone from the inside, I didn't realise how big it was :) Enabling location apps through the web framework.

GS: We get lots of data from carriers around location. 4-5% of people will use an app 15 times on the iPhone. So can we use combinations of location and other date to predict which apps people will use? There is a link between historical location usage and app usage - e.g. for directory applications, someone moving around a lot is more likely to use one.

Q: Where do you get the data to predict churn, age, sex?

GS: Direct from carriers. Usage of the data is tightly regulated, but for an internal purpose like churn measurement is allowed. External stuff like marketing has to be opt-in.

Q: The US is not very regulated when it comes to privacy. Can you give us examples of data extracted for LBS in Europe, compliant with DPA and Privacy laws?

A: (TA) Latitude. (FP) At Nokia we opted against doing a friend-finder product. The issue is tracking vs publishing; Latitude seems a little naive.

RVDK: Location updates are social behaviour - when people update their location manually, they're saying something.

FP: Manual location updates are more interesting and more relevant.

Q: How far can prediction go?

GS: We take a group of people, a year of behaviour, crunch the data, then work out if they e.g. like rap music, go out at night, and can see we've accurately predicted what they like.

FP: I have to make an effort to physically go somewhere - so it has a lot of meaning. Dopplr are doing interesting stuff with location and the social graph; I tried it and they predicted places I'd like with surprising accuracy, based on very little data.

TM: Most advertising is based on your home, which is where you spend very little time.

FP: But this doesn't matter. We used to derive intent from demographic data, now we can predict it from real behaviour.

XCC: In the console world we manually mine data.

Q: The location visualisation and predictions are fascinating. Can you say anything about apps that generated swarm behaviour?

GS: Talks about taxi visualisations; these bring a feedback loop which modify behaviour.

TM: We've not seen anyone build swarm behaviour, but don't have access to the data in realtime so it's hard.

Q: Context is more than location, it's about a network too.
FP: We've seen lots of interesting examples... we can derive more than location from the phone. We form social graphs every day by interacting, the explicit "we are friends" of Facebook is very raw. If we could mine your everyday life we could be more subtle.