Finding proper time intervals for dynamic network extraction

Abstract Extracting a proper dynamic network for modeling time-dependent complex system is an important issue. Building correct model related to finding out critical time points where exhibits considerable change. In this work, we propose measure similarity detect intervals. We develop three metrics, node, link, and neighborhood similarities, any consecutive snapshots of network. Rather than label or user-defined threshold, use statistically expected values proposed similarities under null-model state whether the changes critically. experimented on two different data sets with temporal dynamics: Wi-Fi access logs university campus Enron emails. Results show that, first, reflect similar signal trends topological properties less noisy signals, their scores are scale invariant. Second, generate better signals adjacency correlation optimal noise diversity. Third, using allows us find intervals system, leading extraction non-redundant modeling.