VK has taught recommendation algorithms to take into account the future interests of users
- Новости
- Internet and technology
- VK has taught recommendation algorithms to take into account the future interests of users
AI VK Research researchers have proposed a new approach that allows classical recommendation systems to take into account not only the user's current interests, but also the impact of recommendations on his future preferences. This was announced on July 28 by the VK company's press service.
The developed approach is already being used to solve individual application problems and was adopted at the Customer Journey workshop of the KDD 2026 international conference. The solution can be integrated into existing systems without completely redesigning their architecture.
Alexander Dyakonov, head of the AI Research Department at VK Research, noted that today many studies are focused on large neural network models, but in many services classical algorithms remain the most effective solution.
"We have shown that they can be supplemented with a planning mechanism: take into account not only the current recommendation, but also its possible impact on the user's further interaction with the content. This approach improves the quality of recommendations without significantly increasing computing costs," said Dyakonov.
The new approach was tested on several open datasets, including VK-LSVD. In offline experiments, he showed a higher quality of recommendations compared to the classical method. In the future, the team plans to explore ways to better account for user feedback and adapt the approach for broader industrial applications.
On July 16, VK agreed to sell 100% of the RuStore app store to Dmitry Pankrushev, CEO of the developer company. It was noted that after the transaction, the store's team will continue to work as before, focusing on the development of functionality, product updates and service stability for all categories of users.
Переведено сервисом «Яндекс Переводчик»
