Protecting users’ privacy online

Protecting users’ privacy online

Ten internet searches can be enough to reveal your identity online, according to research from Macquarie University and CSIRO’s Data61 that was presented at The Web Conference 2018 yesterday.

Dali Kaafar presented his research at The Web Conference 2018

But the researchers have developed a new method—called Incognito—to better protect users’ online privacy through obfuscating the web data they leave behind.

“Every time we go online to surf the web or share information, we leave behind a trail of our personal data, interests and intentions,” explains Professor Dali Kaafar, Scientific Director of the Optus Macquarie University Cyber Security Hub.

“This data could be used to reveal private or sensitive information about us, even if our identities are not known and we’re not sharing personally identifiable information.”

Current methods to improve the privacy of our online data fail because they are not comprehensive enough, cannot be applied to all Web data (for example our activity on social media sites) and are not effective against adversarial attacks.

The method Dali’s team has proposed addresses these shortcomings by first better predicting the privacy risk of specific data, and then minimising that risk by obscuring it with lower risk data that carries similar meanings (known as semantically similar data).

The method is particularly powerful as it is also resistant to potential attackers who know everything about the data and the algorithms and parameters in use.

The researchers used two web datasets—AOL users’ search queries and reviews of Android apps on Google Play—to test their method.

They found that it only took 10 search engine queries for the average risk of identifying a user to rise to 100%, whereas using their method this risk could be reduced to 0%.

However this came at a cost in some cases, of losing some of the original meaning of the data. The researchers found that by using their method the utility of the data could be reduced by an average of 22%.

In the future, the team hopes to create an app using their method that users will be able to install as a browser plug-in.

Read the paper.

Published on 27 April 2018.

Back to the top of this page