Peer review has been used in both online and offline classrooms to inspire creativity, gather feedback, and lessen instructor grading loads, especially for design-based tasks without definitive rubrics. To explore the nuances and quality of peer feedback, we developed UX Factor, a peer grading platform that aims to characterize the behavior of peer reviews and the consistency of the ranking models used to aggregate these reviews. This system harnesses the power of pairwise comparisons to minimize bias and encourage context-driven analysis. We adopted UX Factor in a user interface course of 133 students and teaching assistants (TAs) across 3 different individual design projects over a semester and found that the system was effective in eliciting high-quality feedback. We saw that raters have higher agreement than random preferences, and with at least 15 ratings per submission, a simple average of ratings produced rankings that were consistent to both the raw ratings and other more complex models. These rankings were robust to disagreeable raters and changing class sizes, demonstrating the potential of comparative peer review to match the quality of expert feedback at scale.
https://doi.org/10.1145/3479863
The 24th ACM Conference on Computer-Supported Cooperative Work and Social Computing