A Sentinel-Based Peer Assessment Mechanism for Collaborative Learning

This paper introduces a novel mechanism to improve the performance of peer assessment for collaborative learning. Firstly, a small set of assignments which have being pre-scored by the teacher impartially, are introduced as “sentinels”. The reliability of a reviewer can be estimated by the deviation between the sentinels’ scores judged by the reviewers and the impartial scores. Through filtering the inferior reviewers by the reliability, each score can then be subjected into mean value correction and standard deviation correction processes sequentially. Then the optimized mutual score which mitigated the influence of the subjective differences of the reviewers are obtained. We perform our experiments on 200 learners. They are asked to submit their assignments and review each other. In the experiments, the sentinel-based mechanism is compared with several other baseline algorithms. It proves that the proposed mechanism can effectively improve the accuracy of peer assessment, and promote the development of collaborative learning.