Social semantic networks: Measuring topic management in discourse using a pyramid of conceptual recurrence metrics

Studies of conversational encounters typically employ manual discourse analysis methods to reveal participants' topic management patterns, usually focusing on turn-by-turn interactions within a specific social context. These analyses, while powerful, are time-consuming to apply and can prove difficult to generalize. Recurrence analysis has recently been applied to discourse datasets to model how individual terms and concepts recur over whole conversation time scales and relate patterns of recurrence to topic management practices by individuals. In this paper, we propose a new multi-level quantitative method for modelling the topical interaction dynamics in conversation based on conceptual recurrence quantification analysis. The new protocol develops a hierarchy of speakers and their interactions, and partitions recurrence based on these groups. The new protocol is evaluated against expert human coding of television broadcast interviews. Our analysis reveals topic use patterns and networks of conceptual engagement (person-person and group-group) that show experts preferentially engaging with other experts rather than with laypeople, findings that are consistent with prior expectations for this discourse, although never before expressed as metrics. The studies provide a starting point for new computational protocols to provide fast, semi-automated methods for measuring the degree of conceptual interaction between individuals and groups.