Machine learning helps Google find and remove 750,000 bad publishers from ad network

Charlie Spargo's picture
by Charlie Spargo

In 2018, Google worked hard at removing as many adverts, advertisers, and ad-hosting pages classified as "bad", as well as tackling an enormous ad fraud ring alongside the FBI.

Utilising machine learning tools educated through 330 detection classifiers, the search giant took down a total fo 2.3 billon adverts across its network. In 2017, this number was in fact much higher, reaching 3.2 billon.

New scams and fraud methods put Google in an arms race to identify and cut out what are referred to as "bad ads". To name a few, these might include illicit ticket resellers, political ads with undisclosed backers, and bail-bond providers making a profit.

Google was required to introduce 31 new advertising policies last year, and removed 59 million phishing ads alone.

It also worked with the US Department of Justice to expose a group known as 3ve, which is reported to have created three billion bids every single day at its peak, scamming advertisers out of around $36 million over four years.

734,000 publishers and app developers were removed from the ad network, while 1.5 million apps had ads disabled. If a page violates Googles publishing policies on the basis of "misrepresentative, hateful or other low-quality content" the network removes it as a potential host of advertising.