Latent Patterns Detection and Interpretation in Multi-Layer Temporal Network

As a time-varying evolution graph, a Multi-Layer Temporal Network (MLTN) contains the temporal dynamic of multiple subgraphs connected by various types of interactions. Compared to a single-layer network, MLTN implicates more information and complicates relationships, which leads the way that the interaction patterns in MLTN are too complicated to be extracted. In this work, we proposed a compatible MLTN encoding method for converting diverse MLTNs into tensors. Next, based on the tensor factoring technique, a tensor can be factored to main components to help us obtain the main interaction patterns in MLTN. However, the factoring result is hard to understand. Therefore, we built an MLTN visual analytic system called MuNeEye to help system users understand the factoring result of MLTN and further explore the latent patterns of MLTN. Three cases show that the method proposed by us is effective in pattern detection from various types of MLTN, such as social networks, conflict relationships between countries, and computer networks.