|
Abstract:
|
Dyadic data is generated by observed interactions between any two individuals in a population. For data sets with N individuals, modeling who contacts whom, and how often, involves computing N-chose-2 likelihoods. Without making restrictive assumptions, this problem becomes intractable for all but the smallest data sets. We present a framework for modeling dyadic data that manages this scalability problem, while accounting for interdependent variation of unobserved customer traits. By exploiting the discreteness of the Dirichlet process (which we use as a nonparametric Bayesian prior on incharacteristics), we dramatically reduce the number of likelihood computations at each iteration of an MCMC algorithm. We demonstrate the need for, and effectiveness of, our model using a data set of call records from a major Chinese cell phone service provider.
|
