Thanks to the development of machine learning technologies (ML models), postal services manage to minimize the receipt of unwanted emails by users. Maria Anisimova, a programmer-researcher in Mail.ru Mail’s machine learning team, told socialbites.ca how this works.
According to the expert, ML model is a mathematical function created using higher mathematical methods and optimization algorithms to solve problems or predict a specific event based on available data. Simply put, a machine learning model uses algorithms to find and remember patterns in data.
“Antispam systems recognize approximately 100 million spam emails per day. Many people think that spam is easy to detect because it is monotonous and boring, but nowadays spammers come up with different tricks to divert users’ attention. For example, they send letters about easy and quick earnings (“You won your million rubles!”) or phishing (“You urgently need to update your data, follow the link!”). There are many tricks like this, all of them can detect Antispam ML systems. Machine learning algorithms analyze all the information that comes with letters: the sender’s reputation, URL, image and text characteristics, and then detect suspicious patterns, protecting users from possible threats,” Anisimova explained.
Machine learning models are automatically retrained on a regular basis (thanks to user requests and clicks of the “This is spam” button). This is how antispam systems learn to quickly respond to even the latest scammers’ tricks.
“Besides that, do you remember how many times you wanted to unsubscribe from those annoying newsletters that were crowding your inbox? The fact is that not all email sending systems are ready to delist mailboxes, so mail algorithms analyze “trusted” mail every day. If the sender refuses to comply with the user’s requirements, machine learning systems take blocking measures,” explained the expert.
It is also important that two more ML models in the mail work – Automatic Registration and Hack Analysis. Today, Automatic Records’ advanced ML system blocks approximately 130 thousand suspicious records per day. At the same time, to protect already registered users, Hack Analysis systems analyze the actions in the user’s mailbox online and identify hacks within a few seconds after receiving a threat signal.
It is important to understand that by hacking email, attackers want to obtain passwords or links to restore access to the user’s personal data (for example, from social networks, games, bank accounts or popular marketplaces). As soon as the Hacking Detection system sees such suspicious activity, it immediately blocks the mailbox to hide personal data from the attacker. “The most reliable way to prevent hacking is to protect your account with two-factor authentication,” he added.
Russians before saidHow do scammers get phone numbers?