Temporal Network Prediction and Interpretation

Temporal networks refer to like physical contact whose topology changes over time. Predicting future temporal network is crucial e.g., forecast the epidemics. Existing prediction methods are either relatively accurate but black-box, or white-box less accurate. The lack of interpretable and motivates us explore what intrinsic properties/mechanisms facilitate networks. We use learning algorithms, Lasso Regression Random Forest, predict, based on current activities (i.e., connected not) all links, activity each link at next time step. From coefficients learned from algorithm, we construct backbone that presents influence links in determining link's activity. Analysis backbone, its relation series aggregated reflects which properties captured by algorithms. Via six real-world networks, find step a particular mainly influenced (a) (b) strongly correlated close distance (in hops) network.