Triadic balance and network evolution: Predictive modeling with signed networks

Abstract This paper introduces a novel approach for identifying dynamic triadic trans- formation processes, and it is applied to three networks: two directed and one undirected. Our method significantly enhances the prediction accuracy of net- work ties. While balance theory offers insights into evolving patterns of triadic structures, its effects on overall network dynamics remain underexplored. Exist- ing research often neglects the interaction between micro-level balancing mech- anisms and overall network behavior. To bridge this gap, we develop a method for detecting dynamic triadic structures in signed networks, categorizing triangle transformations over two consecutive periods into formation and breakage. We analyze the impact of these structures on temporal network evolution by incor- porating them into exponential random graph models across three networks of varying size, density, and directionality. To address the complexity of multi-layer networks derived from signed networks, we modify the temporal exponential ran- dom graph model framework. Our method significantly improves out-of-sample prediction accuracy for network ties, with additional predictive power from incor- porating negative network information. These findings highlight the importance of considering the triadic transformation processes of balance triangles in studying temporal networks, warranting further research.