Fachdidaktische Qualitätssicherung von Machine-Learning-Modellen in der Hochschuldidaktik

Individuelles KI-gestütztes Feedback im Videoanalyse-Tool ViviAn

Abstract

This article provides an insight into the development of AI models for the video analysis tool ViviAn, which is accompanied by subject-specific didactic analyses for quality assurence. As a result, the interaction of four machine learning models trained with a student self-label approach is presented. The analyses conclude that “self-labeling” is sufficient for low-inferential labeling processes to create AI models that come close to didactic expert ratings. However, this presupposes that sufficient subject data is available for the training of the models.