Google has developed a new filtering system Spam called RETVec (Resilient & Efficient Text Vectorizer). It is ” one of the biggest defense upgrades in recent years », assures the company in a blog post.
Scam emails in most cases have a title that differs from legitimate emails. Malicious actors often use text manipulation to fool spam filtering systems. This includes homoglyphs (characters that look like letters), emojis and even keywords. Although they are easily identifiable with the human eye, the task is more complex for computer tools. With its new device, Google promises greatly improved efficiency in filtering spam in Gmail.
Convincing results
RETVec is based on a machine learning TensorFlow model. It is able to analyze visual similarity to identify the meaning of words rather than their actual content. Over the past year, Google has been testing it internally “ to evaluate its usefulness and found it to be very effective for security and anti-abuse applications “.
The results were convincing. “ Replacing the previous text vector of the Gmail spam classifier with RETVec allowed us to improve the spam detection rate by 38% compared to the baseline and reduce the false positive rate by 19.4%. », Details the Mountain View firm.
The system also reduces computational costs
Another advantage of the system is that it allows “ significantly reduce computational costs ”, while taking care of “ all languages and all UTF-8 characters without requiring text preprocessing “. Indeed, the model has 200,000 parameters, compared to millions for the others. This makes it light enough to run on a local device.
Google has now deployed it to Gmail. RETVec is available as open source on GitHub, to allow other developers to integrate it as a text classifier.
Download
8
- Free, unlimited email storage
- Advanced email filtering and easy searching
- Contact and calendar management integrated into the interface
Gmail is a free email service offered by Google. It allows people to send and receive emails online, and offers a range of features such as email searching, contact and calendar management, and interface customization.
Gmail is a free email service offered by Google. It allows people to send and receive emails online, and offers a range of features such as email searching, contact and calendar management, and interface customization.
Sources: Google, Ars Technica
10