The Yandex team has implemented its most advanced ARGUS recommendation technology, which runs on generative models, into its proprietary advertising system. Now, neural networks take into account users' actions and interests on the internet over a year, instead of several dozen days as before.
As the company emphasized, this is the largest update to Yandex's neural network advertising architecture, allowing advertisers to target advertising more accurately and users to receive more relevant offers.
ARGUS (AutoRegressive Generative User Sequential modeling) technology is based on generative models and combines dozens of neural networks, each of which solves its own task: predicts the probability of a click, assesses the probability of conversion, or takes into account the context of user interaction with advertising. The amount of memory has increased 30 times — from 256 to 8000 events, which allows for better analysis of anonymized data and selection of advertising.
Yandex Advertising processes up to 1.2 million requests per second, and about 10 billion events occur daily within the system. ARGUS is already used in key company services such as Yandex Music, Market, and Lavka.