Online Behavioural targeted Advertising (OBA) has risen in prominence as a method to increase the effectiveness of online advertising. OBA operates by associating tags or labels to users based on their online activity and then using these labels to target them. This rise has been accompanied by privacy concerns from researchers, regulators and the press.
Providing advanced solutions to audit Online Behavioural Advertising and its misuses is fundamental to contribute to a more transparent and privacy-preserving online advertising ecosystem.
In this line, a recent study conducted by TYPES researchers, presents a novel methodology for measuring and understanding OBA in the online advertising market. The authors show a novel methodology based on training artificial online personas representing behavioural traits like ‘cooking’, ‘movies’, ‘motor sports’, etc. Using this methodology as basis, they have built a measurement system that is automated, scalable and supports testing of multiple configurations.
Using this methodology, a first campaign of measurements has been performed. The analysis of the obtained data reveals that:
(1) OBA is a common practice, 88% of the analyzed personas get targeted ads associated to all the keywords that define their behavioural trait. Moreover, half of the analyzed personas receive between 26-62% of ads associated to OBA.
(2) The level of OBA attracted by different personas shows a strong correlation with the value of those personas in the online advertising market (estimated by the CPC suggested bid for each persona).
(3) We provide strong evidences that show that the online advertising market targets behavioural traits associated to sensitive topics related to health, politics or sexual orientation. Such tracking is illegal in several countries. Specifically, 10 to 40% of the ads shown to half of the 21 personas configured with a sensitive behavioural trait correspond to OBA ads.
(4) We repeat our experiments in both US and Spain and do not observe any significant geographical bias in the utilization of OBA. Indeed, the median difference in the fraction of observed OBA ads by the considered personas in US and Spain is 2.5%.
(5) We repeat our experiments by having first set the Do-Not-Track (DNT) flag on our browser and do not observe any remarkable difference in the amount of OBA received with and without DNT enabled. This leads us to conclude that support for DNT has not yet been implemented by most ad networks and sites.
The intention of the authors with this work is to pave the way for developing a robust and scalable methodology and supporting toolsets for detecting interest-based targeting. By doing so TYPES team hopes to improve the transparency around this important issue and protect the advertising ecosystem from falling into the Tragedy of the Commons paradigm previously discussed in this blog.