Tell me who your friends are and I’ll tell you who you are, as the saying goes. Google has apparently taken that age-old adage to heart in trying to protect users from less than conscientious Android apps. Such apps might use permissions and data more than they should, at least compared to other similar apps that behave properly. And while the reasoning behind such peer grouping, Google unsurprisingly injects a pinch of machine learning to make it a less overwhelming task.
Whereas Apple reviews each and every app that goes into its app store, Google automates the process and relies on machine learning instead. While more efficient, some malicious apps do end up slipping through the cracks. That’s especially true for apps that really do look and behave innocently. Save for the fact that they might be getting your location or using data when they don’t have any business to.
To identify such crafty apps, Google resorts to peer group analysis. In a nutshell, it compares each app with other apps like it to see if it is doing anything out of the ordinary. If most coloring apps don’t use location permissions, an app that tries to get permission to read the GPS is going to raise red flags. Users are usually asked to approve or deny permissions, but they don’t usually know why an app would need such. It’s definitely more effective to cut the app off before it can even be installed.
But comparing every app against its peer is more work than what’s possible for humans. In fact, even just grouping them can be tedious enough. Sure, you can group them by the general categories you can find them in, like games, productivity, social, etc. But those are too broad and too massive to be usable in this context. This is where that oh so useful machine learning comes in. Google analyzes and groups apps based on a variety of vectors, including app metadata, the app’s text description, and even user installs.
Such an automated peer grouping and peer group analysis system is definitely a smart way of dealing with a gargantuan task in the most efficient way possible, something Google has traditionally been good at. However, it hasn’t exactly been good at keeping such intrusive or even downright malicious apps at bay. Hopefully the wealth of data and machine learning improvements will help improve the situation. At least before some big, crippling bug hits Android users hard.