Enhancing MarkEd with Peer Feedback Functionality with LLM Support

Peer feedback can improve learning by encouraging deep engagement with course content, exploration of diverse perspectives, and critical thinking. For computing and software engineering students, the practice of peer feedback can contribute to future employability. However, incorporating peer feedback mechanisms into courses can be difficult due to problems such as reduced student motivation and trust, and the resulting low-quality feedback leads to dissatisfaction. This paper presents an extension to [ToolZ], a marking and moderation tool developed at the [redacted university], to facilitate high-quality peer feedback. Once a course organiser sets up peer assessment, the system gradually takes students through the steps of making a submission and providing feedback in the form of anonymous inline comments to a number of their peers. As students write their feedback, it provides them with guidance powered by a Large Language Model (LLM) on improving its actionability, clarity, and constructive tone, while discouraging them from over-relying on such guidance. Moderation by human markers is also introduced to validate peer feedback correctness. The enhanced system was trialled as part of formative assessment in a large second-year introductory software engineering course. Feedback from students and markers indicated a positive impact on learning, with most students reporting that the LLM guidance helped them reflect on their comments and learn about what constitutes high-quality feedback. Markers and a lecturer indicated a reduction in workload compared to other systems, and all groups rated the system’s usability highly, although some improvements could still be made.