Machine-Learning basierte Analyse von latenten Profilen des physikdidaktischen Wissens
摘要
Personal Pedagogical Content Knowledge (pPCK) is a crucial component of the professional knowledge of (prospective) teachers. Empirical research has assessed the development and influencing factors of pPCK and shown pPCK’s significance for professional knowledge and the quality of professional actions. For more detailed research on the connections between pPCK and performance in prototypical practice situations, more differentiated empirically based descriptions of the internal structure of pPCK are necessary. However, such approaches have mostly been based primarily on theoretical-normative considerations or have been limited to hierarchical perspectives. This article therefore presents an approach for the non-hierarchical, data-based description of latent competence profiles of pPCK based on a Computational-Grounded-Theory-inspired workflow. First, a latent profile analysis for the investigation of latent pPCK-profiles with a focus on cognitive requirements as empirically separable subscales of pPCK is carried out using a dataset of 846 responses to the physics pPCK test instrument from the ProfiLe-P+ project with a predominantly open response format. Subsequently tendencies in the language use of the pPCK-Profiles are investigated using Topic Modeling. To confirm the results in terms of the Computational Grounded Theory with evidence of latent connections between the respondents’ test answers and the pPCK-profile assignments, a Machine-Learning-based system for the automated assignment of test responses to the pPCK-profiles is then created. Four latent pPCK-profiles with a non-hierarchical character emerge. Finally, the development of test subjects in the context of these pPCK-profiles is presented in a longitudinal analysis. The results offer opportunities for future investigations of the relationships between individual pPCK components and, for example, the quality of teachers’ action on a more detailed level.