Guiando la intervención del profesorado en la evaluación por pares explotando un modelo gráfico probabilístico

Abstract

Among the active teaching-learning methodologies, formative assessment technique of peer assessment, where students assess each other, stands out. Although this technique could be used to grade students too, there are doubts about the reliability of the ratings given by peers. Under the hypothesis that peer assessment can be modeled to efficiently guide teachers in choosing which activities to correct, the objective is to obtain a reliable estimate of the grade of the tests that the teacher has not reviewed. Probabilistic graphical models are used for modeling, together with a Bayesian machine learning method that adjusts to peer and teacher ratings. A procedure is proposed that suggests, one by one, which work the teacher should grade next to reduce the uncertainty in the model. The teacher decides how many activities to grade based on their own criterion of tolerable uncertainty. This proposal, validated in real data, shows promising results and has the potential impact of helping to extend peer evaluation as an evaluation and grading technique, reducing doubts among teachers about the reliability of the grade estimates.