Yandex has created the Sona neural network: music recommendations have become more accurate and personal
Yandex has introduced Sona, a new-generation unified generative neural network for its recommendation service. Unlike traditional systems where dozens of separate programs handle music search and selection, the new model solves the entire chain of tasks independently.
The main innovation of Sona lies in combining track search and sorting into one smart system. The neural network assigns unique digital codes to audio recordings based on their sound and people's preferences. By analyzing listening history, skips, and ratings, the algorithm instantly selects the most suitable tracks from millions of available options.
The technology has already passed successful tests on smart speakers with Alice. Compared to the previous multi-stage system, Sona showed impressive results:
• total user listening time increased by 6.3%;
• the number of positive ratings (likes) increased by 11.4%;
• the size of the service's active audience rose by 4.5%.
Yandex specialists continue to test and refine the algorithm. In the future, the company plans to adapt the Sona technology for other services, including product recommendations and advertising platforms.
ORIENT
