Russian Researchers Make Graph Neural Networks Several Times Faster
Researchers at AI VK Research have developed a method that makes it possible to use graph neural networks with extremely large datasets of interconnected information. Their paper has been accepted for presentation at the international CIKM 2026 conference.
A graph represents data as a collection of objects and the connections between them. In a social network, for example, the objects are users and the connections represent friendships. Graph neural networks analyze these relationships and are used in recommendation systems, social networks and other applications.
The new approach can scale graph neural networks to hundreds of millions of objects and billions of connections. It also reduces memory requirements and speeds up data preparation for training. The researchers tested the method on an anonymized social graph containing 194 million users and 28 billion connections. It required more than 100 times less memory and cut data preparation time by a factor of 2.5. In VKontakte’s friend recommendations, the number of friend additions increased by 16%, while the number of users who added a recommended friend rose by 11.5%.
