What You Say In The Conversation Affects The Flow: Modeling Conversational Flow Using Nlp Methods

Conversational flow, defined as the experience of smooth, efficient, and mutually engaging conversation, is an important factor that shapes people’s feelings about social relationships and interpersonal impressions. Given these significant social outcomes, however, we do not yet know how the content of the conversation predicts the flow. The current research used NLP algorithms on a rich conversation dataset collected from speed networking events (487 dyads) to investigate whether and how conversational flow can be predicted by what people say in the conversation. We conducted novel text analyses on the conversation transcripts and identified meaningful linguistic features that can reliably predict how a person's flow is evaluated by their conversation partner. For example, both positive and negative emotions predict higher flow, and features like filler pauses, yes or no questions, and short answers predict the lower flow. We also found that features that appear at the beginning of the conversation contain strong signals that can predict the flow of the entire conversation. Finally, we built a machine-learning model with linguistic features that predicts flow at a higher accuracy than benchmarks like word count. Together, this research demonstrates that conversational flow is a direct function of the conversational choices people make, and we suggest clear interventions for people to improve conversational flow and create better interpersonal outcomes.