(archive site)

Global Inference & Online Privacy

Global Inference & Online Privacy

Understanding and Managing the Impact of Global Inference on Online Privacy is a scientific study of the implications of global inference for privacy across online social networks. User-supplied content — photos, videos, and text — is a crucial ingredient in many websites and services today. However, many of the users who provide content do not realize that their uploads may be leaking personal information, because the form of that information can be difficult to grasp intuitively. In this project (also called GeoTube), we are exploring the ways in which information in public posts in different venues, each of which may seem innocuous in itself, can be correlated to draw inferences that could be used to the poster’s detriment.

For example, we have demonstrated how it would be possible for a potential attacker to use geotags (GPS data) attached to posts, timestamps, and writing style to determine that accounts on Yelp, Flickr, and Twitter belong to the same person — and in some of the experiments, correlating those features was even more indicative than comparing usernames. This means that someone who was trying to maintain separation between their identities on different sites could be found out, for example, by tweeting from a restaurant and then posting a review of it on Yelp. Similarly, we have also demonstrated that it is possible to use speaker-recognition technology to pick out Flickr videos that were posted by the same person, by examining the audio tracks.

Global Inference & Online Privacy is in part an outgrowth of our research on multimodal location estimation, in which we use visual and acoustic features and textual tags to estimate the recording location for user-generated media content that does not have GPS data attached. Some of our current work is looking at how multimodal location estimation — a growing field — could feed into inference processes and privacy attacks. We are also looking at how the information collected by online data brokers (companies that aggregate personal data about consumers) affects privacy.


Our experiments with using information from uploader-supplied tags to geolocate videos — and determine whether the uploader was away from home — won a Best Poster award at the 2012 Korean Student Technical & Leadership Conference.

Project Results

Global Inference & Online Privacy in the News:

Our work on cybercasing, or using GPS-tagged or localizable images to infer the current location of a person or a valuable object, has been featured in major national news outlets:

Global Inference & Online Privacy Publications


Understanding and Managing the Impact of Global Inference on Online Privacy is a joint effort between ICSI’s Audio & Multimedia and Networking & Security groups and researchers at University of California – Berkeley.

Researchers @ ICSI:

Collaborators @ UC Berkeley:

  • Giulia Fanti
  • Michael Tschantz


Understanding and Managing the Impact of Global Inference on Online Privacy is funded by National Science Foundation grant CNS-1065240. The opinions, findings, and conclusions described on this website are those of the researchers and do not necessarily reflect the views of the funders.