Applied Mathematics Seminar Event
For more information regarding this event, please email Ben Macarthur at B.D.Macarthur@soton.ac.uk .
Event details
Continuous-time temporal networks with tie decay In this talk I will present recent work on modelling networks from streams of interactions in continuous time. In these networks the weight of an edge represents the strength of a tie between two nodes. Our modelling framework makes a distinction between interactions (discrete events at specific times) and ties (the strength of a relationship, which evolves in time). The two key modelling assumptions of this framework are: 1) In the absence of interactions a tie between two nodes decays exponentially with rate alpha. 2)When the nodes interact the tie experiences a discontinuous jump. The decay rate represents the deterioration rate of relationships and can be tuned to reflect a variety of scenarios. Alternatively, tie decay can represent the diminishing value of old information. We introduce an adaptation of the popular PageRank centrality score to tie-decay networks, and provide an efficient algorithm to compute it in streaming data applications. We illustrate these networks on synthetic examples, and a dataset of tweets about the NHS. Finally, we will discuss potential avenues for future research and other applications, such as recommendation systems