Identifying Stable States of Large Signed Graphs

Abstract

Signed network graphs provide a way to model complex relationships and interdependencies between entities: negative edges allow for a deeper study of social dynamics. One approach to achieving balance in a network is to model the sources of conflict through structural balance. Current methods focus on computing the frustration index or finding the largest balanced clique, but these do not account for multiple ways to reach a consensus or scale well for large, sparse networks. In this paper, we propose an expansion of the frustration cloud computation and compare various tree-sampling algorithms that can discover a high number of diverse balanced states. Then, we compute and compare the frequencies of balanced states produced by each. Finally, we investigate these techniques’ impact on the consensus feature space.