Social networks contain clusters (clouds) of nodes centered in the "elephants" nodes which degrees exceed sufficiently high thresholds and they are surrounded by "mice" nodes. To model clusters, their mean size and a minimal time to reach a large node we obtain the extremal index of the node degree. The latter reflects the dependence structure of the process and approximates the reciprocal of the mean cluster size and the limit distribution of a maximal node degree in the sample. Processes generated by PageRank and random walk sampling techniques are considered. Related joint distributions of node selection are obtained.