Building an Effective Representation for Dynamic Networks

A dynamic network is a special type of network which is comprised of connected transactors whichhave repeated evolving interaction. Data on large dynamic networks such as telecommunications networksand the Internet are pervasive. However, representing dynamic networks in a manner that is conduciveto efficient large-scale analysis is a challenge. In this paper, we represent dynamic graphs using a datastructure introduced by Cortes et. a]. [Q]. We advocate their representation because it accounts forthe evolution of relationships between transactors through time, mitigates noise at the local transactorlevel, and allows for the removal of stale relationships. Our work improves on their heuristic argumentsby formalizing the representation with three tunable parameters. In doing this, we develop a genericframework for evaluating and tuning any dynamic graph. We show that the storage saving approximationsinvolved in the representation do not affect predictive performance, and typically improve it. We motivateour approach using a fraud detection example from the telecommunications industry, and demonstratethat we can outperform published results on the fraud detection task. In addition, we present preliminaryanalysis on web logs and email networks