Online friendships light up shadow social networks

FRIENDING someone on Facebook makes an association public, but many relationships are never professed online. Now there's a way to use the structure of an online social network to deduce the offline connections, dubbed "shadow connections", between people who don't use the service.

The technique may alarm those concerned about online privacy - but it could also be applied to other network types, helping to reveal hidden brain connections or biochemical pathways.

Previous research has shown it is possible to deduce information about members of online social networks that they have not explicitly revealed - such as their location, personality or sexual orientation - from what their friends reveal online. But these predictions all require a person to have set up an online profile.

By contrast, Fred Hamprecht, of Heidelberg University in Germany, and his colleagues were curious about what it is possible to discover about the relationships between people who do not belong to an online social network. Privacy advocates have already expressed concerns about this but no one has attempted to quantify it mathematically.

As shadow connections are by their nature unknown, Hamprecht's team started by creating a model of an offline social network interlinked with an online one (see diagram). The researchers took data on the friendships between tens of thousands of Facebook users collected from five college campuses in 2005. The team labelled a subset from each college "members". The rest were "non-members" and treated within the model as if they did not have an online profile.

Next the researchers calculated which non-members were likely to have appeared in the email address books of which members and added this to the model. This was to mimic the fact that social network users typically disclose their email address books when they sign up.

Using the network structure of four of the university campuses, a machine-learning program picked out attributes that seemed to predict whether two non-members knew each other, such as how many members knew both of them and how many knew one but not the other. The program used only the relationships between members and the email data, both of which a social network company could access.

When the researchers then used the program to predict links between non-members in the remaining college Facebook network, 40 per cent of the predictions were correct. By contrast, they calculate that using a random guessing approach, just 2 per cent of suggested connections would be right (PLoS One, DOI: 10.1371/journal.pone.0034740).

Team member Katharina Zweig, of the Technical University of Kaiserlautern, Germany, says social network users might want to know how data they submit could be used to reveal information about others.

She adds, however, that "we do not say that social network platforms are actually doing this". Facebook declined to comment, though last October a student alleged that Facebook Ireland was creating shadow profiles of non-members. A later privacy audit found no evidence for this.

The ability to exploit the interconnectedness of known parts of a network to learn about unknown parts could help reveal hidden connections between neurons in the brain, or biochemical pathways, says network theorist Jon Kleinberg of Cornell University in Ithaca, New York. "In many domains, we're seeing data that gives us a detailed glimpse into one part of a much larger network."

by MacGregor Campbell
http://www.newscientist.com/l