Modelling large or big networks with exponential random graph models
Time: 14:00 - 15:00
Venue: Room 1.218, University Place, 176 Oxford Rd, University of Manchester, Manchester, M13 9QQ
Sorry, this event has now ended.
Speaker: Dr Johan Koskinen, University of Manchester.
The Data Science Institute looks forward to welcoming Dr Johan Koskinen (School of Social Sciences, University of Manchester) at our next Advanced Data Analytics Seminar.
Abstract: Exponential random graph (ERG) models is a family of log-linear distributions for network ties that is rapidly becoming increasingly popular in applied social network analysis. The sufficient statistics of the ERG model can be derived out of a simple set of dependence assumptions that prescribe in what ways the ties of a network may depend on each other. This gives rise to a model that both explicitly models dependencies in the interactions of nodes as well as models that have statistics that in and of themselves have meaningful interpretations. These (local) dependencies do however mean that ERG models do not marginalise and the models hence do not scale up. In this talk I will introduce ERG and their basic ingredients and discuss issues of scaling and how this affects both modelling and inference. In the process I hope to dispel of some misconceptions about the properties of ERG models. I will provide a few examples of approaches for estimating ERG models for large and potentially Big networks.