Request For Research Presentations For the PrivacyCon Conference
We would like to present our recent results on informing users on demographic inference based on data commonly shared with different services: their location patterns. Consumers are routinely asked to share location information in order to personalize and improve commercial services. But rarely are they aware of the inferring power of that data, which remains hard to gauge and act upon. This raises concerns for two reasons: some amount of location data may be sufficient to reveal answers to questions such as "Are you female?" or "Do you belong to a minority ethnic group?" which consumers would not be willing to disclose willingly. Perhaps even more concerning is the fact that those answers can be revealed without even the users being aware of it. Ideally, users must be informed before disclosing their footprints of its potential inferring power. But answering that question accurately is rendered difficult by a lack of training data due to its sensitive nature, and the subtlety of location patterns that may emerge. In this project, we show that those challenges can be overcome. Leveraging collections of photosharing networks and simple inference tools, we are able to determine that a small amount of coarse grain location data suffice to accurately infer ethnicity even with simple algorithms. We will then present an end user tool allowing one to connect personal accounts from Instagram, foursquare and Twitter, and learn how her location history shared through digital footprints may reveal their demographics.