Inferring Dynamic Dependency with Applications to Link Analysis
Michael R. Siracusa, John W. Fisher · 2006
Statistical approaches to modeling dynamics and clustering data are well studied research areas. This paper considers a special class of such problems in which one is presented with multiple data streams and wishes to infer their interaction as it evolves over time. This problem is viewed as one of inference on a class of models in which interaction is described by changing dependency structures, i.e. the presence or absence of edges in a graphical model, but for which the full set of parameters are not available. The application domain of dynamic link analysis as applied to tracked object behavior is explored. An approximate inference method is presented along with empirical results demonstrating its performance.