RUTUBE has launched its own multimodal ML model for video classification according to the IAB standard. It simultaneously analyzes the video sequence, audio track, title, and description of the video, after which it determines its topic with a stated accuracy of over 93%.
The new system is significantly more detailed than the previous one: the number of categories has increased by approximately 10 times. For example, within the "Auto" section alone, more than 16 directions can now be identified – from repairs and spare parts to motorcycles and electric transport.
Initially, the markup was connected to advertising algorithms to more accurately select the context for impressions. Later, it is planned to be used in recommendations and search – the system will be able to find the right videos even without keywords from the author.
Currently, recommendations already account for more than half of the views on RUTUBE. In the first half of 2026, the average monthly audience of the service exceeded 82.3 million people, and the daily audience grew by 18% – to 20.4 million.
Read more on the topic:
- Views on Rutube soared by almost 20%: viewers spent 1.9 billion hours on the platform
- Gazprom-Media Holding CEO Zharov: Rutube has blown up the video hosting market
- Video, AI, and 2,000 Servers: RUTUBE to Build Its Own Data Center