NEW YORK, USA: There’s growing evidence that if you are part of a social network, the structure of the network itself can reveal important information about you, regardless of what you have published yourself.
For example, in December we looked at a study of a computer gaming network, which showed that if you have friends who cheat at computer games, you are much more likely to be a cheat yourself or to become a cheat in the near future.
In a way, that makes sense. We are much more likely to copy the behaviours of friends than of other people.
Today, Kazem Jahanbakhsh and pals at the University of Victoria in Canada add an interesting corollary to this work.
These guys have studied the geographical clusters of photos that users upload to Flickr, the popular picture sharing website. The task they set themselves is to determine an individual’s home town looking only at the geotags of photographs they have uploaded.
It’s no surprise that people take most of their photographs near their home. But they also take photographs in clusters at other locations such as holiday destinations and such like. That makes the problem of estiamting the home location a little more difficult. The trick that Jahanbakhsh and pals solve is to find an algorithm that can separate the home location from the other clusters.
As it turns out, many users publish their home location on their Flickr profiles, so their algorithm is straightforward to check. Jahanbakhsh and co say that sure enough, it guesses reasonably well. "In 70% of the cases our algorithm has predicted the place of living of people with low error," they say.
Jahanbakhsh and co go on to make the claim that the reverse process is also useful: that it’s possible to make a good guess about the location of a photograph given the user’s hometown.
That’s seems a step too far. It’s certainly likely that a single photograph in a user’s collection will have been taken in their hometown but it’s hard to see how they could say anything about the other locations a user might visit. Of course, there may be some useful correlation to be exploited by looking at the locations of a user’s friends. But Jahanbakhsh and co make no claim about this.
The important point is that these kinds of algorithms can piece together many individual tidbits of information to build up a remarkably accurate picture about an individuals. That makes it hard to keep track of privacy.
So there’s almost certainly a market out there for an application that monitors the drip feed of data that an individual releases.
Worth keeping an eye on.