Who thinks who knows who? Socio-Cognitive Analysis of an Email Network

Interpersonal interaction plays an important role in organizational dynamics, and understanding these interaction networks is key for any organization, since they can be tapped to facilitate various organizational processes. The principal roadblock to studying organizational networks, however, has been the difficulty in collecting data about them. The approach of conducting surveys/interviews is fraught with issues of scalability, logistics, and reporting bias – especially since a survey of this nature can seem quite intrusive. Widespread use of computer networks for organizational communication provides a unique opportunity to eliminate all these difficulties and automatically map the organizational network to a high degree of accuracy and detail. This paper describes an approach to automatically build organizational networks by tapping into the e-mail server, which observes all organizational communication. Specifically, our approach focuses on studying how communication between actors is perceived by other actors in a social network. Such type of networks are formally referred to as socio-cognitive networks (i.e. a "who knows, who knows who" network). The key issues addressed by this paper are the representation and construction of a socio-cognitive network from electronic communication data as well as identifying and proposing techniques analyses of such networks. Each actor in the socio-cognitive network has a "set of beliefs" about the communication between other actors in the network. We propose a model for representing and constructing the sociocognitive network, where communication between actors is represented as probability distributions. Each actor attempts to estimate these distributions using Bayesian inference, based only on the communication he/she observes. The conflict in beliefs between different actors, and their divergence from reality, is analyzed by comparing the belief models of actors. Measures which quantify the notions of "misperception" (divergence of an actor’s beliefs from reality), "agreement" (similarity in beliefs of different actors) and "consensus" (general agreement in a group of actors) are introduced. We present results for these on the Enron email corpus. Our work, in addition to providing novel computational schemes to sociologists, also highlights an important aspect of social network analysis research where one focuses on how actors' perceptions evolve over time.