Mining social networks to predict your app choices

 

UNSURE which app to download to your smartphone? While you waver, internet giants like Apple and Google could soon be predicting what you will install by analysing how you interact with your friends.

A team at MIT's Media Lab in Cambridge, Massachusetts, analysed social-network behaviour among smartphone users to see if it could help forecast app downloads. The research could give developers a valuable insight into why someone chooses to download particular apps.

"We thought, 'Can we use social networks to find which apps people might find interesting?' " says team member Wei Pan. "We found it difficult."

The team gave smartphones running Google's Android software to 55 postgraduate students living in university accommodation. Each phone recorded masses of social interactions from several sources over a period of five months: logging phone calls, detecting physical proximity using Bluetooth and keeping track of activity on Facebook.

They also logged when each student installed a new app from the more than 30,000 available on the Android Market app store.

They created software that analysed these data in three stages. First, it aimed to identify a user's most significant friends. It then noted the apps those friends were using. Finally, it worked out the probability that a given app is owned by a user, based on what their closest friends owned.

Judging the importance of each friend proved a tricky task. "We ended up wanting to compute the optimum social network, to describe exactly who were your closest friends," says Pan. But data taken from a single source didn't offer enough insight. So their software combined all the sources to predict a list of apps that a particular user had installed from the 821 different ones downloaded by the entire group. On average across their tests, 45 per cent of the apps on each person's list was predicted correctly. In comparison, complete guesses were only about 10 per cent accurate (arxiv.org/abs/1106.0359).

The work will be presented at an artificial intelligence conference in San Francisco in August.

The technology could one day be rolled into any app, allowing developers to analyse social network data from a user's smartphone and shift their marketing focus accordingly. "They're reinforcing that we don't have one social network - it's the combined effect of networks that's important," says Bernie Hogan of the Oxford Internet Institute, UK. "These guys are at the forefront of figuring out what to do with them."

 

refer to:http://www.newscientist.com/article/mg21128195.600-mining-social-networks-to-predict-your-app-choices.html