​Google's machine learning helping it catch spam to Gmail

Google has shed some light on why Gmail users should hardly ever see spam in their inbox, and almost never see wanted email in the spam folder.
Written by Liam Tung, Contributing Writer

Google says less than 0.1 percent of email in the average Gmail inbox is spam while wanted mail in spam folders is under 0.05 percent.

Those stats are, according to Google, the result of using its artificial neural network to sift through billions of incoming emails to weed out unwanted messages and phishing attacks.

Were Google not doing everything it could to weed out spam, Gmail would likely be unusable. According to security firm Kaspersky, 59.2 percent of all email it filtered in the first quarter of 2015 was spam, with its senders jumping on newly-released domains, such as .work and .science, to sneak past spam filters and deliver advertisements or malware.

From Gmail's earliest days, Google been using machine learning to improve its spam filter, relying on its 900 million users to flag up unwanted messages through the services "report spam" and "not spam" buttons. However, as it noted in a blog post today, users still occasionally have to click the "not spam" button, which essentially meant that they had to wade through their spam folder to find an email that was wanted, but flagged as spam -- for example, a monthly statement alert from the bank.

To tackle this issue, Google is releasing a new system called Gmail Postmaster Tools that will allow those companies that send email in bulk to analyse data on delivery errors, spam reports, and reputation. While the diagnostic toolset is aimed at companies, the benefit for individual Gmail users should be, according to Google, "no more dumpster diving".

Alongside these new tools, Google shed some light on how its artificial neural network technology, which powers products like Google Search and Google Now, is helping to tackle the "especially sneaky spam".

Google doesn't provide any precise stats on how well its system is at detecting phishing email, but it says it's the best it's been at rooting out email impersonation.

"Thanks to new machine learning signals, Gmail can now figure out whether a message actually came from its sender, and keep bogus email at bay," it said.

Lastly, Google is using machine learning in its spam filter to understand different tastes between Gmail users.

Read more

Editorial standards