Darshat Shah

Principal Software Eng Manager

Ad Quality Year in Review 2015

Online advertising plays a big role in keeping many online services low cost or free of charge. This is a win for both users and website owners, however, it can also attract questionable actors, who are constantly looking for ways to push their low quality or even dangerous content to unsuspecting users, convincing them to take action they may not have intended. In really malicious cases, bad ads may cause users to lose personal information, or infect their devices. Protection is needed to ensure bad ads are weeded out and remaining ads are safe, useful and relevant.

At Bing Ads, we take the safety and security of our search users very seriously. We start with an extensive team of domain experts who are up-to-date with latest attack patterns, as well as an operations team that is constantly refining policy enforcement to keep the ecosystem clean from new threats and compliant with laws in 40+ countries and multiple languages. In addition, we incorporate escalations from users and advertisers into policy enforcement improvements. Coupled with an extensive backend infrastructure, we work 24/7 to ensure ads that are shown are policy compliant and safe for our users. Each and every ad entering the system is scanned using automation powered by machine learning and then manually reviewed when necessary to ensure correct policy decisions.

Ad Quality Ecosystem

Attack Patterns

In 2015, we rejected hundreds of millions of ads, and identified thousands of sites pushed by bad actors for being in direct violation of our policies. Bad actors often engage in misleading behaviors with the intent to mislead and potentially damage our users. Some of the models we detected include:

  • Unrealistic weight loss claims and dieting schemes
  • Quick money making schemes and work from home jobs
  • Questionable deals on luxury items like designer purses

We initiated stricter checks on advertising that offers toll free numbers to fix computer problems – many users searching for computer solutions are not tech savvy which makes them extra susceptible to misleading and dangerous ads. There have even been times our machine learning algorithms have flagged accounts that look innocent at first glance – e.g. a dealer specializing in kids’ furniture – but on close examination we find malicious intent. The backend machinery runs 24/7 and uses hundreds of attributes to look for patterns which help spot suspicious ads among billions of genuine ones.

In addition, we’ve improved our handling of dead links which helps both our users and advertisers. In some cases, advertisers may violate our policies unintentionally, and we’ve improved messaging on steps they need to take to bring individual ads back into compliance.

Looking Back

Below is a summary of our enforcement actions in 2015.

Ad Quality Lookback

Safety First

Digital domain moves at the speed of thought. As we fight current threats, new and more sophisticated ones continue to emerge. We are continuously updating our policies, threat models, and Machine Learning based detection mechanisms in order to fight bad ads faster and with better accuracy. User experience on Bing is critical for our platform, and we are determined to ensure search ads on Bing are useful and safe.

Keep reading