Method for generating graphs with control of statistical properties

In this paper we propose a method of conditional graph generation that takes into account statistical characteristics of graphs. These characteristics are divided into two groups. The first group, called simple features, can be computed by efficient deterministic algorithms with complexity not more than quadratic of the number of vertices. This is dictated by the costliness of using computationally complex algorithms on graphs that are close to real graphs in size. The second group of features is generated in the hidden space and is responsible for graph regularities that cannot be described by «simple features». This approach allows to generate graphs with precisely defined statistical characteristics, while preserving their diversity. Moreover, this method can be applied to generate graphs with similar structure to the original one. The performance of the proposed method is confirmed by a computational experiment conducted on the Citeseer and Cora datasets.