Content-Aware Ad Blocking

We train our machine learning model on a set of labels generated by existing ad-blocking filter lists, so our baseline is accuracy at reproducing the hand-curated filters. Performance during day-to-day use may sometimes exceed that of existing ad blockers by extrapolating from known cases in ways that hand-written filters are unable to do. Results below are from running our adblocker on 60 websites and 157 URLs (an admittedely small sample size) and comparing to existing adblockers. Our adblock performs with far better recall and similar precision to existing adblockers.