Predicting the Evolution of Social Networks: Optimal Time Window Size for Increased Accuracy

摘要

This study investigates the data preparation processudfor predictive modelling of the evolution of complex networkedudsystems, using an e–mail based social network as an example. Inudparticular, we focus on the selection of optimal time windowudsize for building a time series of network snapshots, whichudforms the input of chosen predictive models. We formulate thisudissue as a constrained multi–objective optimization problem,udwhere the constraints are specific to a particular application andudpredictive algorithm used. The optimization process is guidedudby the proposed Windows Incoherence Measures, defined asudaveraged Jensen-Shannon divergences between distributions of audrange of network characteristics for the individual time windowsudand the network covering the whole considered period of time.udThe experiments demonstrate that the informed choice of windowudsize according to the proposed approach allows to boost theudprediction accuracy of all examined prediction algorithms, andudcan also be used for optimally defining the prediction problemsudif some flexibility in their definition is allowed.