Temporal Pattern Mining from Evolving Networks

Recently, evolving networks are becoming a suitable form to model many realworld complex systems, due to their peculiarities to represent the systems and their constituting entities, the interactions between the entities and the timevariability of their structure and properties. Designing computational models able to analyze evolving networks becomes relevant in many applications [1]. The evolution of networks has attracted the interest of the research in Mathematics and Physics [2], which offer theoretic mathematical models based on specific laws and global features (e.g. degree, density, diameter). However, in realworld applications, the networks often do not follow one specific model, they may reflect the combined behavior of several models and may even have exhibit unexpected characteristics. Despite the theoretic frameworks, the research of datadriven approaches is becoming promising thanks to the possibility to characterize the evolution of complex systems by means of techniques originally designed for evolving data. Two main categories of techniques can be recognized on the analysis of the changes, that is, clustering-based and pattern-based. Clustering-based approaches [3,4] focus on the changes of network-based or node-based indicators, thus they lead to discovering changes that regards only the whole network or some nodes, without any information on the topology. Conversely, pattern-based approaches [5,6] rely on the frequent pattern mining framework considering subnetworks, thus they may find valuable changes as they operate on portions of the whole network.