Google bets on machine learning to block spam in Gmail

Google is betting on its TensorFlow machine learning framework to block spam messages in Gmail.

The technology, according to the search giant, has been applied to train additional spam filters for users of its e-mail service. The action paid off, as Google ensures that this extra help was able to block more than 100 million spam every day.

With around 1.5 billion users, Google says that Gmail is already able to block 99.9% of spam every day, but this update is meant to account for the remainder that can obscure our email boxes.

It's not today that Google applies artificial intelligence to trim the edges of Gmail. The company has been working in this sieve for years. But while so-called rule-based filters can block the more obvious, say, spam, TensorFlow looks for new patterns that might suggest that an email is unreliable.

"Using TensorFlow has helped us to block image-based messages, embedded content emails, and messages from newly created domains that try to hide a low volume of spam messages at a legitimate traffic, "wrote Neil Kumaran, Google product manager in a post posted on the company's blog.

By being trained in this way, the algorithms are able to work with a large number of metrics, including from the format of an email to the time it was sent. At the same time, with the application of TensorFlow, it is easier for Google to customize the spam protections for each user. "What one person considers to be spamming may find that message important. Think here about newsletter subscriptions or regular e-mail notifications of some service you subscribe to," Kumaran added.

Related Articles