Scaling ad-filtering technology
A presentation at Chief in Tech Summit 2022 by Jutta Horstmann
Scaling ad-filtering technology For almost a quarter of a billion users (and counting) Jutta Horstmann (COO) | Gertrud Kolb (CTPO) | eyeo GmbH Chief in Tech Summit 2022
Talking points 1 2 3 What is Ad Filtering and why should you care? Scaling ad filtering technology - how we adapt our product portfolio. Basis of it all are Filterlists - how we want to automate.
eyeo… Who?
Most popular ▪ Most popular browser extensions worldwide ▪ Technology available on all major browsers and platforms
The problem User has no control Result Users see more ads Publishers earn less & have to allow more aggressive ads Advertisers don’t stand out & budgets move to more aggressive ads Content becomes less conveniently and openly accessible. 5
The solution User has control Result Users see fewer ads Diverse and valuable content remains accessible as a public resource Cap on ads makes them more valuable for publishers Incentive for advertisers to run user-friendly ads 6
Our mission For the value exchange to have any impact, it must be sustainable, which requires a balance in power. Empowering a balanced and sustainable online value exchange Our solutions give power to the stakeholders. Each side gets something out of the exchange.
Our vision Any value exchange that is balanced between all sides; everyone benefits. Putting you in control of a fair and prosperous internet We aim to create solutions that give all stakeholders choice in how the internet works best for them. The healthier and stronger the ecosystem, the more enduring it will be.
Next at eyeo: 1 billion users
Scaling ad-filtering technology How we adapt our product portfolio
Our product journey to grow our user base to 220m users and further More Users → More Publishers → More Revenue Monthly Active Users is one of our key success metrics. How do we get more users? Direct user Desktop Web Extensions Desktop Browser: Acceptable Ads Opt-Out! Direct user Mobile Partnerships Mobile Partnerships Desktop AdBlock Browser Android Chromium SDK Chromium SDK Extensions Samsung Internet Browser: iOS SDK Javascript SDK FF Extension (out of support) WebView SDK (out of support)
Filterlists & Allowlist the basis of it all How we want to automate
Filterlists & Allowlist Filterlists (Ad-Blocking) ▪ ▪ ▪ Allowlist Chromium-based browser which integrated eyeo’s Chromium SDK ▪ ▪ ▪ Sets of filter rules that automatically remove unwanted content from websites, including annoying ads, bothersome banners and troublesome tracking Regular expressions in text files Ownership: external Filterlists (Open Source) eyeo products uses these filter lists for blocking Contains whitelisted publishers Contains exception rules for acceptable ads Ownership: editing & maintaining by eyeo
Filterlists: facts & figures ▪ ▪ ▪ ▪ ▪ ▪ ▪ Provide information what to block/hide Open source Big community > 40 languages > 500 filter lists Most widespread filter list is the Easylist with - > 80 000 filter rules - > 400 commits per week - > 40 authors After blocking ads, potential whitespaces or ad containers are hidden using element hiding which effectively rearrange the webpage. Websites don’t look broken!
CHALLENGE: a lot manual processes with the Filterlists and our Allowlist Filterlists Ensure quality of Filterlists - Filter rules can break websites! - No prediction possible before rolling out changes on filter lists! - Too conservative vs. too liberal - Eyeo distributes filter lists from external authors and has to ensure the quality! - Helping tools: Monitoring, Crawling, Manual testing Anti-Circumvention - Website operators take action against ad blocking → Circumvention - eyeo takes action against circumvention → anti-circumvention - “Cat-and-mouse-game” - Detection of circumvention based on user-reports & monitoring Allowlist Maintaining Allowlist: - adding / removing publisher and acceptable ads partners Ensure Acceptable Ads Standard: - Checking continuously the most important websites - Complete manually with some helping tools
Project Moonshot: Automation of Ad Filtering VISION We want to filter ads automatically Our Motivation ▪ Reduce human intervention ▪ Improve efficiency with automation ▪ Be the best in ad filtering ▪ Cutting-edge in the ad-blocking space
Automated Ad Filtering via Machine Learning Collecting data & training the model Data Collector Extension DOM / HTML / CSS → Websites data with labeled ads (instead blocking) Extract data nodes & train ML-Models: NLP-based & Graph-based Data Collector Server Crawler crawls a list of websites & executes a Data Collector Extension Automated Ad Filtering using the model AdFiltering Extension AI Filters use ML-Model Extension AdFiltering filters ads based on the model Website with filtered ads
Project Moonshot Journey ahead Pipeline automation Large-scale data collection Learning paradigms exploration Performance evaluation & monitoring 1 8
Summary and next steps 1 Scaling is a journey 2 Automation is crucial 3 Open questions Want to help us grow our vision and fundamentally change how online advertising works? Get in touch!
Thank you. Questions? Jutta Horstmann Gertrud Kolb j.horstmann@eyeo.com @smphr g.kolb@eyeo.com
Scaling ad-filtering technology For almost a quarter of a billion users (and counting)
