Analyzing interaction communication networks in enterprises and identifying hierarchies
With the proliferation of electronic modes of communication (e.g., e-mails, short messages), a group of people in an enterprise can form several distinct Communication Interaction Networks, or CINs for short. A CIN is essentially a graph representation of “who talks to whom” among a group of individuals. In this paper, we conduct an empirical study of two modern enterprises and focus on three main questions: (Q1) how CINs from the two enterprises look, (Q2) how employees use the different available communication modes within an enterprise, and (Q3) how much information we can extract regarding the roles of their participants. We address these questions using empirical CINs from the Enron Corporation and a communication provider, using information from the exchange of e-mails, phone-calls, and short messages (SMS). For Q1, we reveal the following key structural properties that are shared by all the CINs in our study: they have high edge density, high clustering coefficient, and close to zero assortativity coefficient. For Q2, we observe that employees have differences in how they use the various communication modes. This suggests that different CINs capture different behavioral properties within an enterprise. For Q3, we propose HumanRank, a method of ranking individuals based on their importance (e.g., CEOs having higher rank than ordinary employees) using only the interactions between them. Next, using HumanRank, we introduce an unsupervised and parameter-free algorithm that identifies hierarchies by separating managers from ordinary employees. Our algorithm achieves above 70% accuracy and outperforms the state-of-the-art.