A group of researchers from AI VK Research has introduced an innovative approach that enhances the capabilities of graph neural networks when working with large datasets. This new method was successfully tested on friend recommendations within the social media platform VKontakte, leading to a remarkable 16% increase in the number of users adding friends based on these recommendations, as reported by VK representatives.
The approach employs a graph structure to represent data in terms of objects and their interconnections. In a social networking context, for instance, users serve as the objects while the friendships between them represent the connections. Graph neural networks are adept at analyzing these relationships and are widely utilized in recommendation systems, social networks, and various applications requiring the analysis of interconnected data.
The method developed by AI VK Research is versatile and applicable across various industries that deal with extensive networks of related data. It enables the scaling of graph neural networks to handle hundreds of millions of objects and billions of connections, significantly reducing memory requirements and accelerating the data preparation process for training.
During their testing, the researchers utilized an anonymized social graph featuring 194 million users and 28 billion friendship links. They devised a data representation technique that reduced the necessary memory footprint by over 100 times. Additionally, this solution takes into account the evolving nature of social graphs over time, facilitating a 2.5 times faster data preparation for training. The results showed an 11.5% increase in users who added at least one recommended friend during the testing phase.
This breakthrough has significant implications for the market, potentially setting a new standard in how companies handle large-scale data analytics and offering a competitive edge in the realm of recommendation systems. Competitors may need to adapt rapidly to maintain their market position in the face of such advancements.
Informational material. 18+.
