Modeling Response Time in Digital Human Communication

Abstract

Our daily lives increasingly involve interactions with other individuals via different communication channels, such as email, text messaging, and social media. In this paper we focus on the problem of modeling and predicting how long it takes an individual to respond to an incoming communication event, such as receiving an email or a text. In particular, we explore the effect on response times of an individual's temporal pattern of activity, such as circadian and weekly patterns which are typically present in individual data. A probabilistic time-warping approach is used, considering linear time to be a transformation of "effective time,'' where the transformation is a function of an individual's activity rate. We apply this transformation of time to two different types of temporal event models, the first for modeling response times directly, and the second for modeling event times via a Hawkes process. We apply our approach to two different sets of real-world email histories.The experimental results clearly indicate that the transformation-based approach produces systematically better models and predictions compared to simpler methods that ignore circadian and weekly patterns.