AI Can Help Instructors Help Students: An LLM-Supported Approach to Generating Customized Student Reflection Responses

This innovative practice paper presents an LLM-supported technique to help instructors respond effectively to periodic students' reflections. Efficient communication between instructors and students is integral to supporting a productive learning environment. Recognizing the significance of understanding students' perceptions and challenges, we present the initial implementation of a system to help instructors analyze and respond to students' feedback promptly and effectively. This research is inspired by and extends prior works where instructors sent progress check emails to students, with some works finding that such communication increased students' motivation. To collect feedback, we administer regular student reflections throughout the semester that capture how students feel about the course and uncover the challenges they face. This regular feedback-gathering approach allows instructors to better track their students' progress and respond to comments throughout the semester to provide guidance. However, reading and responding to each reflection manually in the context of their overall learning experience can be time consuming. To address this challenge, we introduce an LLM-based automated approach that generates tailored, performance-contextualized responses to student reflections that can be used to guide first-contact interventions. The generated reflection responses (GRRs) address issues discussed in student reflections and provide advice, support, course information, and follow-up questions to the students. Additionally, they provide feedback to students based on their accomplishments and behavioral data within the learning management system (LMS), such as submission patterns. In this work, we discuss our method of generating responses based on students' reflections and their LMS behavior. We also present example scenarios of the proposed approach. Preliminary results indicate that this approach can help instructors facilitate positive educational interactions with students and that the participating students view the interventions favorably, fostering a constructive learning environment. This work provides an initial presentation of our large language model-based response generation method to motivate further investigation into AI-assisted student support mechanisms