Temporally Evolving Models for Dynamic Networks

    The research of complex networks and large graphs generated a wide variety of stochastic graph models that try to capture the properties of these complex systems. Most of the well-known models can describe a static graph extracted from a real-world dataset. They are capable of generating an ensemble of graphs, in which all graph instances are similar in terms of specific statistics to the original one. For example, models that capture the power-law degree distributionempt to model the actual temporal evolution of large graphs. Our goal is to give temporal stochastic graph model for the temporal dynamics of these complex systems.

    Our models address the link prediction problem introduced by Liben-Nowell and Kleinberg, in a \emph{temporal} setting. More specifically, we try to predict accurately each new link in the graph at the time when it is created in the network. This experimental setting is similar to our method introduced for recommender systems. We explain this setup in case of dynamic graphs. For baseline algorithm, we apply online matrix factorization on temporal network data.

    Frederick Ayala, Robert Pálovics, András A. Benczúr
    International Conference on Computational Social Science, Helsinki, June 2015