In this article, the problem of hierarchial matrix clusterization is discussed. For this, the influence of individuals on the community was used. The problem of dividing the community into groups of related participants has been solved, an appropriate algorithm for finding the most influential community agents has been proposed. Clustering was carried out using an algorithm for reducing the adjacency matrix of a directed graph with nodes representing members of a social network and edges representing relationships between them. The applications to the problems of working groups, advertising in social networks and complex technical systems are considered.