Technology

Yandex develops generative model Sona for recommendation system

Illustration: Tezkun / AI. Not a photograph of the event.

Yandex has introduced its own generative artificial intelligence model, Sona, designed for content search and ranking.

The development was carried out by the company's R&D team alongside engineers from the Alisa and Yandex Music services. The new architecture unites several stages of recommendation selection into a single module, eliminating the need to tune dozens of individual algorithms and manually select features. The system independently analyzes user actions and then generates the final output.

Music recommendations in smart speakers with Alisa became the first service to test the technology. According to the company, during A/B testing with the complete replacement of the previous chain of algorithms, the listening time of recommended tracks grew by 6.3%, the number of likes increased by 11.42%, and the number of requests to repeat a track rose by almost 18%. The share of active listeners showed an increase of 4.5% during the week of the experiment.

To select audio tracks, the model uses composite Semantic ID codes that take into account track metadata, the first 90 seconds of playback, and audience reactions. For the initial training of the algorithm, the company applied a separate "teacher" model trained on a year-long array of user actions. In operational mode, the service receives an updated version of Sona every 10 minutes based on recent listens and ratings.

Published 1 October 2026, 18:02 · No updates · Russian original

Discussion
  1. Loading comments…
No links or insults. One comment per 20 seconds.

Comments on this story live under its post in our Telegram channel. Comment on Telegram