This is the summary of the article written by Toubiana, et al., which proposes a method of online behavioral advertising which doesn’t compromise user privacy. You can get the PDF of the behavioral targeting article here: Adnostic..
One of the main problems of behavioral targeting is that it seems to be inherently conflicting with privacy. In this paper, a group of students propose an alternative around the dilemma. They created a system called Adnostic, and they discuss how the advertiser, publisher and ad-network can keep their advertising methods while accepting this proposed system as a compliment to their methods.
Goal of Adnostic
The goal of Adnostic is to allow ad-networks to serve the most significant ads without using private information from users. Ad-networks that use Adnostic are still able to provide targeted ads without knowing user clickstream (list of URLs user visited), behavioral profile (user’s interests inferred from clickstream), and ad impression history (ads displayed to the user). Adnostic’s advertising strategy is more limited compared to behavioral advertising, which is not as concerned with privacy. However, the success of this proposed method can be seen on its incentives to ad-networks who wish to support it.
Some incentives of using Adnostic
First incentive is for privacy-conscious publishers. Adnostic is a compliment to the existing behavioral targeting mechanism. This method creates a second option aside from the tracking-based system. That option is the privacy-preserving system. There are sites such as Alcoholics Anonymous, who want to preserve privacy and not allow targeted ads. However, with Adnostic, these types of websites will finally allow targeted ads. Second incentive is for small ad-networks. These ad-networks can’t see the user’s visited pages unless he/she visits sites connected to the ad-network. Small ad-networks can do better with Adnostic, which analyzes user behavior at the browser. Another incentive is that users that opt-out of ads will still receive targeted ads from Adnostic, which doesn’t track their actions. Furthermore, targeting will also be allowed in private browsing mode, because Adnostic doesn’t update user behavioral profile in this mode anyway.
Behavioral Profiling
The Adnostic system locally processes history database to know user’s interests. Each page in the history is classified using natural language processing heuristics. This method can also measure intent to purchase and influence. Furthermore, the problem in traditional behavioral targeting when it comes to computers with more than one user is eliminated. A user is distinguished in this case through keystroke dynamics or analysis of immediate last few pages viewed.
Ad Insertion
When a user visits a web page, this page then points to an ad-network. The content served by the ad-network then identifies if the user has Adnostic or not. If not, the ad-network loads a specific ad to the publisher page. If yes, the ad-network sends n number of ads. Using browser’s history database, Adnostic will choose one of the n ads. Choice of n is configurable. Larger n means more precise targeting but increases network bandwith, and vice versa.