Using semi-automated feedback and assessment for mathematical proficiency: a quasi-experimental study on student learning effectiveness, reliability and time savings.

Feedback is the most powerful engine of any learning process. In the field of Mathematics Education, the possibilities to assess automatedly are therefore being thoroughly explored. However, students face difficulties expressing themselves mathematically on a computer and learning systems can often only assess the outcome and not the solving method. Research indicates that automated tests focus too much on procedural fluency, at the expense of mathematical thinking questions. It takes much effort to develop digital tests and teachers are sceptical of using automated assessments, meaning that paper-and-pencil tests still dominate the math class. One of the characteristics of mathematical assessment is that wrong answers tend to exhibit patterns among the student population. Consequently, teachers often repeat their feedback and marks. This brings us to the idea of semi-automated feedback and assessment: by correcting handwritten tasks digitally, feedback can be saved and re-used. This could lead to more elaborate feedback and interesting time savings, but also opens up possibilities to extensively monitor the individual learning process of students, and to apply adaptive differentiated instruction using Bayesian networks. A Bayesian network is a probabilistic graphical model of a student's proficiency. We want to focus on the learning gains that semi-automated evaluation systems can offer, but also explore the reliability, time savings and acceptance levels of such systems.

Keywords:ARTIFICIAL INTELLIGENCE (AI)

Disciplines:General mathematics, Didactics of school subjects, Educational technology, Teacher training