Prompt Learning With Multiperspective Cues for Emotional Support Conversation Systems

The emotional support conversation (ESC) system is tailored to reduce the distress of individuals who are experiencing emotional challenges. To achieve this goal, the system must fully understand the emotional state of the user and employ appropriate strategies to provide effective emotional support. However, most existing methods predominantly focus on the user’s psychological state and the causes of user’s emotional problem, neglecting key cues, e.g., topics of conversation and listener’s psychological state. This leads to an incomplete understanding of the user’s situation and ineffective emotional support. This article presents a novel method that utilizes <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</u>rompt learning with <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">m</u>ulti<underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</u>erspective <underline xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">c</u>ues (PMPC) to generate emotional support responses. Specifically, we extract multiperspective cues from the dialogue history and the causes of user’s emotional problem. To fully utilize the different cues within the ESC system, we categorize them into two groups: <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">semantic enhancement cues</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">semantic constraint cues</i>. Subsequently, we construct two prompts based on different categories of cues, which are designed to thoroughly understand user’s emotional predicament and generate an engaging support response, respectively. Experimental results on the ESConv dataset demonstrate that our proposed PMPC can surpass other approaches in both automatic and human evaluation metrics, providing compelling evidence of its efficacy and potential impact in real-world applications.