Finding How We Get Recommended Ads and MoreCategory: Science & Technology
Posted: August 19, 2014 11:40AM
Unless you are exceptionally careful about what information is on the Internet, there is a good chance you have been presented with a recommendation based on your online activities. Of course we all know that services and websites collect information from emails, video views, and product views, but how exactly do they generate the recommendations? That is what researchers at Columbia University want to know, and so they have developed XRay to provide greater transparency on the Internet.
Approaching the problem of how our information is used is tricky, because much of the Internet operates like a black box. Without the ability to view the processes involved in generating the recommendations, XRay has to rely on black-box correlations between inputs and outputs. At first the researchers worked with theoretical results, which were encouraging, but only theoretical, so they soon started running experiments on Gmail, Amazon, and YouTube and refining the design. Eventually XRay achieved complete success with each experiment, matching theoretical predictions in complex cases, which suggests it can scale up well.
Though the current system has only been run on Gmail, Amazon, and YouTube, it should be service-agnostic, so any site that tracks you could be studied with XRay. Thus far it has revealed that it is possible to target sensitive topics and that there does appear to be abuse of the recommendation systems. You can see examples results at XRay's website: XRay: Transparency for the Web.
Source: Columbia University