Mutual information: inferring tie strength and proximity in bipartite social network data with non-metric associations

The current study was a first step exploration of a new method that used mutual information-based measures to represent tie strength or proximity between individuals from bipartite social network data with non-metric associations. Unlike network datasets with explicit links between nodes, bipartite networks provide only implicit indications of the probable existence of connections. Therefore, as a measure of the amount of information shared between these two random variables, mutual information can be used to infer social network structure in bipartite network data. A literature review found surprisingly low utilization of mutual information in social network analysis, although it was widely used in other areas of network analysis. Two studies in the current thesis showed that mutual information can be effectively used to infer tie strength and proximity from bipartite social network data with non-metric associations. Other social network analysis techniques such as graph theory-based centrality measures and hierarchical cluster analysis can then be applied to the mutual information-based measures to further investigate the underlying social network structure such as detecting members of subgroups and detecting important nodes that centered the network. Advantages and potential disadvantages of using mutual information-based measures in social network analysis and future directions in ways of improving this method were discussed