Project Moonshot: Scaling ad filtering

Using machine learning to automate online ad filtering

At eyeo we are dedicated to bringing sustainable balance to the internet. For this to work long-term, content creators, publishers and advertisers must be able to fairly monetize but also must do so without compromising the user experience.

We think the best way to reach this happy medium is with ad filtering. This inspired us to look into using advanced artificial intelligence (AI) such as machine learning (ML) to tackle one of the main issues for internet users today, how to handle intrusive and inappropriate ads which can invade privacy and disrupt the overall user experience. We set ourselves one huge, crazy goal - a moonshot - to create a minimum viable product (MVP) that implements ML to automate ad detection.

In this series, we will take you through our journey to the moon as we look to revolutionize the way ad detection works, and we’ll go over our efforts, the learnings, the challenges and achievements that have arisen during Project Moonshot.

Machine learning (ML), the branch of artificial intelligence (AI) used to solve practical problems, focuses on the idea that machines can learn from collected data by identifying patterns and then making decisions based on that experience. This ability to adapt and see patterns from huge amounts of data can be far more effective and efficient than any individual human or group of humans. With potentially unlimited applications out there, how could we most productively use this emerging technology to give our solutions that competitive advantage and make it the best on the market? 

Good, better, best

At eyeo, we've always been inspired to find ways to serve all internet users. When we started ten years ago, we had a good idea of ad blocking as a tool for user empowerment and as a way to improve how the internet was working but the downsides quickly became apparent for the long term. Without advertising to fund it, quality content would begin to become accessible only through paywalls and subscriptions, which is something not feasible for everyone. 

So we came up with a better idea: ad-filtering which allows advertisers, publishers and creators the ability to monetize their content while facilitating an optimized user experience. Our technology powers products that give users the choice to see advertising that meets an independently researched and maintained set of criteria (think format, size and placement) known as the Acceptable Ads Standard. 

Now the question we posed for ourselves was: how can we take a good idea that we already improved and make it the best? How can we do this more accurately, more effectively and quicker than what is already out there? Our answer: automate online ad filtering to effectively eliminate disruptive and annoying advertisements that interfere with an ideal online experience. 

Practical problems

For our ad-filtering technology to work, we use filter lists that are subject to circumvention and require human development and maintenance. By automating ad detection, we hope to filter the more intrusive and disruptive ads while reducing human intervention of filter list creation and maintenance. We can also optimize ad blocking and ad filtering on mobile platforms and speed up our anti-circumvention efforts. All of which enhances our technology and creates a better product for our partners to distribute to their users. As AI advances, circumvention and bad actors are using it for their own nefarious means. We need to level the playing field by using machine learning to preserve the online experience, user choice and privacy. 

Our journey to the moon

Working on something to serve the whole internet is, in itself, a longshot. Using machine learning to automate the wily and tricky forms of online advertising - now that’s shooting all the way to the moon. Thus Project Moonshot was born. 

In future articles in this series, we will be covering how we deployed this idea, the challenges that arose in the process and talk to our experts about the results and insights they’ve had during this journey. We’ll also be updating, in real-time, new strategies we develop based on our successes and failures.