Google deploys new vector algorithm for Gmail, which can greatly improve spam / phishing email recognition rate.

Thanks to CTOnews.com netizens OC_Formula, Alejandro86, soft media users 1520111, baa ocean for the delivery of clues! CTOnews.com, December 3, according to foreign media PC mag, Google has quietly added a spam filter for Gmail, claiming to have a better performance in identifying spam and phishing emails.

According to reports, this new spam filter is based on the new open source multilingual text vectorizer RETVec (flexible and efficient text vectorizer) on Google Colab. It can map words to vectors or numbers, so it can be used to improve the recognition rate of spam and phishing messages and reduce the false alarm rate.

As cyber criminals continue to create content that bypasses the defense system, current text classification models still have difficulties in identifying fraud and phishing attacks. Google claims that RETVec is trained to detect and understand "character-level" operations, including inserts, deletions, spelling errors and homonyms, while reducing computing costs.

According to Google's own statistics, after applying RETVec to Gmail, the spam detection rate increased by 38% compared with the previous filter, the false alarm rate decreased by 19.4%, and the utilization rate of tensor processing unit (TPU) decreased by 83.13%.

Google claims that models trained with RETVec show faster reasoning speed, and smaller models can reduce computing costs and latency, which is critical for models on large-scale systems and devices.

CTOnews.com Note: the project is now open source on GitHub. Click here to jump.