Event embedding for temporal networks: scientific paper presents new method weg2vec

An interacting group of people, the collectively active neurons in the brain, or the transportation system of a city. These are examples of complex systems, which are all intrinsically dynamical. They can be commonly interpreted as a set of entities, which interact over time and form a network structure coding the architecture of the system in hand. However, conventional network embedding models are often developed for static structures: they consider nodes only and they are seriously challenged when the network is varying in time.

In a paper published in Nature Scientific Reports, ISI Foundation Senior Research  Scientist Laetitia Gauvin, ISI Researcher Maddalena Torricelli and ISI  Fellow Márton Karsai (Central European University) propose a new method of event embedding of temporal networks, called weg2vec (“weighted event graph to vector”), which builds on temporal and structural similarities of events to learn a low dimensional representation of a  temporal network.

This projection successfully captures latent structures and similarities between events involving different nodes at different times and provides ways to predict the final outcome of spreading processes unfolding on the temporal structure. Researchers say this technique can be also used as an online method taking into account temporal events on the run.  Moreover, in future works this embedding method may be applied to solve questions such as the detection of key events in misinformation spreading.

“weg2vec:  Event embedding for temporal networks”, Maddalena Torricelli, Márton Karsai and Laetitia Gauvin. Nature Scientific Reports, 28th April 2020, link