Fig. 17
From: Evolving network representation learning based on random walks
Comparative analysis of different strategies for determining when to obtain a network representation. The PERIODIC methods will obtain a new representation every 50 or 100 time steps (i.e., network changes). Our proposed method, ADAPTIVE, is combining a peak detection method and a cumulative changes cut-off method to determine the time to obtain a new network representation. As a result it is able to make more informed decisions and perform better. This is depicted by smaller (on average) changes of the \(RW_{t_{old}}^{t}\), which implies that a more accurate network representation is available for down-stream network mining tasks