Modeling Human Intelligence in Customer-Agent Conversation Using Fine-Grained Dialogue Acts

Smart service chatbot, aiming to provide efficient, reliable and natural customer service, has grown rapidly in recent years. The understanding of human-agent conversation, especially modeling the conversational behavior, is essential to enhance the machine intelligence during the customer-chatbot interaction. However, there is a gap between qualitative behavior description and the corresponding technical application. In this paper, we developed a novel fine-grained dialogue act framework specific to smartphone customer service to tackle this problem. First of all, following a data-driven process, we defined a two-level classification to capture the most common conversational behavior during smartphone customer service such as affirm, deny, gratitude etc., and verified it by tagging chatlog generated by human agent. Then, using this framework, we designed a series of technically feasible dialogue policies to output human-like response. As an example, we realized a smart service chatbot for a smartphone customer using the dialogue-act-based policy. Finally, a user study was conducted to verify its efficiency and naturalness. Since the dialogue acts are meaningful abstraction of conversational behavior, the dialogue-act-based chatbot could be more explainable and flexible than the end-to-end solution.

Duke Scholars

Author Yucheng Jin DKU Faculty