Monitoring Network Changes in Social Media

Econometricians are increasingly working with high-dimensional networks and their dynamics. Econometricians, however, often confronted unforeseen changes in network In this article, we develop a method the corresponding algorithm for monitoring dynamic networks. We characterize two types of changes, edge-initiated node-initiated, to feature complexity The proposed approach accounts three potential challenges analysis First, objects causing standard statistical tools suffer from curse dimensionality. Second, any social likely driven by few nodes or edges network. Third, many applications such as connectedness its centrality, it will be more practically applicable detect change an online fashion than offline version. detection at each time point projects entire onto low-dimensional vector taking sparsity into account, then sequentially detects comparing consecutive estimates optimal projection direction. As long is sizeable persistent, projected vectors converge one, leading jump sine angle distance between them. A therefore declared. Strong theoretical guarantees on both false alarm rate delays derived sub-Gaussian setting, even under spatial temporal dependence data stream. Numerical studies application media messages support effectiveness our method.