Detecting Relational States in Online Social Networks

The state of relationships between actors ion the internet is constantly changing and fluctuating to a social system of constant shocks. Link prediction, community detection, recommendation systems were built from around this fundamentally unstable system. Stable relational states - which hold important and latent deterministic knowledge have often been overlooked in this regard. In this paper, we propose a novel method of quantifying and detecting stability in the relationship between a given pair of actors. Our main algorithm (MVVA) establishes relational stability from a multivariate, autoregressive link feature dynamics perspective. Under our experimental design, we provide another built-in module based on the Hamiltonian Monte Carlo technique to provide a comprehensive cross-validation on the performance and accuracy of our proposed MVVA model.