Sampling on bipartite networks: a comparative analysis of eight crawling methods

Abstract

Sampling networks via crawling has become a feasible and widely used approach when the global network information is dicult to obtain. But there is little focus on two-mode networks, i.e. bipartite networks in which nodes can be divided into two disjoint partitions. In this paper, we adopt eight popular crawling methods (BFS, DFS, FFS, RW, SNS, MHRW, MDRW and RDS) from studies of one-mode networks and evaluate their applicability and performance on bipartite networks. Simulation results show that Metropolis-Hastings random walk (MHRW), maximum-degree random walk (MDRW) and respondent-driven sampling (RDS) perform better than the other methods, and population estimates from them are minimally aected by the structures of degree distribution, number of nodes in two node layers, degree correlation and communities. In addition, we find that strategies used in the sampling designselection approaches for seed nodes, the number of seed nodes, and the number of branches-have very little influence on the estimation bias. Finally, we list suggestions for the selection of crawling methods on bipartite networks under dierent situations.