UC Irvine

Advertising is prevalent across different platforms, especially on the web. To combat this, millions of users rely on privacy-enhancing technologies, such as adblockers. They are powered by filter rules, which are string-based patterns that block and hide advertising and tracking. However, these filter rules are manually curated and continuously maintained by human experts. This is further exacerbated by technical reasons, as advertising and tracking are employed in different ways across websites. In addition, for economic reasons, such as when websites and advertisers attempt to circumvent adblockers to earn revenue through advertising, effectively causing an arms race.

In this thesis, we examine this arms race and develop methodologies and frameworks to improve adblockers in terms of automation, scalability, and robustness. To achieve these goals, we make the following contributions. First, we examine the human effort necessary to update filter rules that combat adblock circumvention by conducting a longitudinal analysis. To detect circumvention, we build CV-Inspector, a machine-learning approach that leverages differential analysis to capture features of circumvention across HTTP and HTML DOM modalities. CV-Inspector reduces the human maintenance cost for filter rules, as it removes the manual monitoring of websites for circumvention. Our second contribution studies the problem of filter rule generation from scratch by developing a reinforcement learning framework called AutoFR. Our formulation enables us to automate the human process of filter rule creation and maintenance. Notably, the AutoFR framework is tunable (e.g., users can explore different reward functions) and applicable beyond ads and the web (e.g., generate rules that block tracking for mobile apps). We demonstrate that both CV-Inspector and AutoFR are effective in a controlled setting and in the wild, i.e., when applied to real websites, even for those we have not trained on. We envision our tools and methodologies will be useful to the adblocking community to improve and automate the creation and maintenance of adblocking filter rules.