Constructing social networks from semi-structured chat-log data

Chat-log data is a little used resource for analyzing human communication in social networks. Some statements in this data do not include the intended username of a receiver or any variant of it, and thus are termed “misaddressed statements”. Constructing social networks from such a semi-structured data and subsequent analyzing require a reliable process to make sure that the social network representation is as truthful as possible. Due to the large size of data, human assignment of statements to receivers is prohibitive. In this paper, we present and evaluate different methods to reliably predict a receiver for these misaddressed statements. We use a set of prediction rules which follow human communication behavior in a group chat and we show their success in constructing social networks.