Enhancing academic stress assessment through self-disclosure chatbots: effects on engagement, accuracy, and self-reflection
Minyoung Park, Sidney Fels, Kyoungwon Seo
Abstract
Abstract Academic stress significantly affects students’ well-being and academic performance, highlighting the need for more effective assessment methods to guide targeted interventions. This study investigates how self-disclosure chatbots—designed to share relevant experiences and thoughts—can enhance academic stress assessments by increasing student engagement, improving accuracy, and fostering deeper self-reflection. Two chatbot conditions were developed: a self-disclosure (SD) chatbot that used personal narratives to build empathy, and a non-self-disclosure (NSD) chatbot. In a randomized experiment with 50 university students, participants interacted with either the SD or NSD chatbot. Results showed that the SD chatbot elicited significantly higher engagement, as evidenced by longer session lengths (15.55 ± 5.92 min) and higher word counts (240 ± 114.02 words), compared to the NSD chatbot (11.31 ± 5.21 min; 162.38 ± 66.24 words). Assessment accuracy—evaluated by comparing results from the SISCO Inventory of Academic Stress with chatbot-generated evaluations—was slightly higher for the SD chatbot (0.936) than for the NSD chatbot (0.862), based on accuracy within a ± one-point deviation. Moreover, students who interacted with the SD chatbot reported deeper self-reflection and developed more actionable strategies for managing their stress. Overall, these findings illuminate the value of self-disclosure in chatbot-based assessments and highlight broader applications for addressing academic stress and mental health challenges in educational settings.