USE OF GRAPH NEURAL NETWORKS FOR SOLVING THE PROBLEM OF CLASSIFICATION OF GRAPH VERTICES

Vasiliev Roman Alexandrovich

The tasks of classifying objects are encountered quite often in the framework of work in the industry. In most cases, they are solved on the basis of basic characteristics (by classical machine learning algorithms), without taking into account graph information. The hypothesis of the study was that methods that take into account the graph structure will give higher quality models. In this paper, several approaches to solving the classification problem using graph information are presented and tested. They are based on real data from a large Russian telecommunications company. Such methods have not been used in the Russian telecom before. Ultimately, in 2 out of 3 studied projects, it was possible to obtain increases in the quality of the models.

Keywords:Neural networks, graph information, graph vertices