Thursday 25 Oct 2012: Learning Hidden Network Structures
Dr Iead Rezek - University of Oxford
Harrison 203 14:00-15:00
Most social, biological, and technological networks display substantial non-trivial topological features, with patterns of connection between their elements that are neither purely regular nor purely random. The topology of most studied networks can be readily extracted from the data, such as the parties communicating via email. However, when such structural information is not available other techniques have to be developed. In this talk I will present an approach for extracting the social structure from point process data. The approach identifies temporal regions of dense agent activity, and extracts links between individuals based on co-occurances. The statistical significance of these connections is also tested using a bespoke permutation test and overlapping network communities are extracted using statistical matrix decomposition. Experiments are performed two large-scale datasets (1 Mio observations) and I illustrate the method's usefulness by examining the temporal dynamics of pairing behaviour, a process that was previously very hard to observe.