Modeling human dialogue-the case of group communications in trunked mobile telephony

In this report we study the dynamics characteristics of conversations among several participants, analyzing trunked mobile telephony databases. We find that the duration of the dialogue sessions deviates systematically from the predictions of Poisson statistics, following probability distributions with either log-normal or power law tail. We propose that such a behavior can be described using the Bouchaud-Mezard (B-M) stochastic model, which suggests that the two probability distributions correspond to different types of human networking in conversations. We discuss the implications of our results in risk assessment in communications.