Grading Open-Ended Questions in an Educational Setting, via Non-exclusive Peer Evaluation

A framework to allow (semi-)automated grading of answers to open-ended questions (“open answers”) is presented. The grading is done by using both the peer (students) assessment and the teacher’s evaluation of a subset of the answers. The Web of data, associated with peers’ and teacher’s assessments, is represented by a Bayesian network (BN). The students are modeled by their Knowledge and by the effectiveness of their evaluations (J). The answer grades in the network are represented as variables, with value in an estimated probability distribution. Grades are updated by evidence propagation and triggered by teacher’s/peer’s evaluation. The framework is implemented in the OpenAnswer Web system. We report on experiments and discuss the effectiveness of the approach.