An Automated Approach to Ad Tracker Detection and Classification

Abstract

In this paper we propose an automated system for detecting, categorizing, and verifying ad trackers on the web. We base this system on the OpenWPM platform developed by Englehardt et al., which we leverage to create an “aggressive” attempt at maximizing tracker coverage while minimizing the potential negative impact on functionality. We explored the suitability of structural “A/B” DOM tree variations in fitting supervised learning models. In doing so, we observe potential advanced tracking methods like cookie syncing in the wild and attempt to explain the limitations of relying on patterns in DOM structural data in classification. Finally, we propose next steps towards the improvement of tracker detection and classification in hopes of overcoming the observed limitations of DOM structural features in generalizing to the diverse content found across the web.