On Sampling Type Distribution from Heterogeneous Social Networks

Abstract

Social network analysis has drawn the attention of many researchers recently. As the advance of communication technologies, the scale of social networks grows rapidly. To capture the characteristics of very large social networks, graph sampling is an important approach that does not require visiting the entire network. Prior studies on graph sampling focused on preserving the properties such as degree distribution and clustering coefficient of a homogeneous graph, where each node and edge is treated equally. However, a node in a social network usually has its own attribute indicating a specific group membership or type. For example, people are of different races or nationalities. The link between individuals from the same or different types can thus be classified to intra- and inter-connections. Therefore, it is important whether a sampling method can preserve the node and link type distribution of the heterogeneous social networks. In this paper, we formally address this issue. Moreover, we apply five algorithms to the real Twitter data sets to evaluate their performance. The results show that respondent-driven sampling works well even if the sample sizes are small while random node sampling works best only under large sample sizes.