Adaptive Dialogue Management for Conversational Information Elicitation

Harshita Sahijwani

Abstract: Information elicitation conversations, for example, when a medical professional asks about a patient's history or a sales agent tries to understand their client's preferences, often start with a set of routine questions. The interviewer asks a predetermined set of questions conversationally, adapting them to the unique characteristics and context of an individual. Multiple-choice questionnaires are commonly used as a screening tool before the client sees the professional for more efficient information elicitation [5]. However, recent proof-of-concept studies show that users are more likely to report their symptoms to an embodied conversational agent (ECA) than on a pen-and-paper survey [3], and rate ECAs highly on user experience [4]. Chatbots allow the user to give free-form responses and ask clarification questions instead of having to interpret and choose from a list of given options. They can also keep the user engaged by sharing relevant information and offering empathetic acknowledgments when appropriate. However, many of the technical challenges involved in building such a conversational agent remain unsolved.