Modeling the evolution of interaction behavior in social networks: A dynamic relational event approach for real-time analysis

There has been an increasing interest in understanding how social networks evolve over time. The study of network dynamics is often based on modeling the transition of a (small) number of snapshots of the network observations. The approach however is not suitable for analyzing networks of event streams where edges are constantly changing in frequency, strength, sentiment, or type in real time.

In this paper, we present a relational event model that directly models interaction rates as a function of endogenous and exogenous variables. The effects of these variables are likely to be dynamic over time themselves. To properly capture the dynamic nature of the network drivers, we develop a moving window approach that can accomodate arbitrarily long memory lengths. We show how Bayes factors and posterior model probabilities can be used to quantify the statistical evidence in the data for the existence, direction, and relative strength of the network drivers, over time.

We illustrate the approach by analyzing streams of email messages about innovation activities between employees in a consultancy firm. We allow the drivers of these email exchanges to vary over time and consider two different memory lengths (60 days versus 150 days).

The proposed methodology can be used to uncover new insights about interaction dynamics in real time event networks. For instance, the illustrative analyses reveal how similarity (in terms of expertise, tenure, or geographic location) affects email network dynamics and how this gradually changes over a year.