On Link Stability Detection for Online Social Networks

Link stability detection has been an important and long-standing problem within the link prediction domain. However, it has often been overlooked as being trivial and has not been adequately dealt with in link prediction. In this paper, we present an innovative method: Multi-Variate Vector Autoregression (MVVA) analysis to determine link stability. Our method adopts link dynamics to establish stability confidence scores within a clique sized model structure observed over a period of 30 days. Our method also improves detection accuracy and representation of stable links through a user-friendly interactive interface. In addition, a good accuracy to performance trade-off in our method is achieved through the use of Random Walk Monte Carlo estimates. Experiments with Facebook datasets reveal that our method performs better than traditional univariate methods for stability identification in online social networks.