From Metrics to Meaning: Large Language Models and the Computational Turn in Embodied Educational Research

Daniel Autenrieth¹Nina Autenrieth²Danyal Farsani^3*

• ¹Rheinisch-Westfalische Technische Hochschule Aachen, Aachen, Germany
• ²Padagogische Hochschule Schwabisch Gmund, Schwäbisch Gmünd, Germany
• ³Norwegian University of Science and Technology, Trondheim, Norway

Educational science has long grappled with a methodological tension: quantitative metrics offer scale but often lack depth, while qualitative inquiry offers depth but is difficult to scale. Mixed-methods approaches have sought to address this tension by combining both paradigms, yet practical constraints of time, labor, and analytical capacity have typically limited how fully researchers can integrate interpretive depth with large-scale analysis. The emergence of Large Language Models (LLMs) and multimodal Artificial Intelligence (AI) provides new opportunities to extend this integration. This article outlines three specific computational methodologies applicable to educational research: Semantic Similarity Rating (SSR), AI-based Qualitative Content Analysis (AI-QCA), and Computational Ethnography via multimodal video analysis. We examine how these methods allow researchers to operationalize hermeneutic processes, such as interpreting student reflections or analyzing classroom habitus, at a scale previously reserved for standardized testing. By detailing the theoretical basis, opportunities, and limitations of these approaches, we argue for a "computational turn" in educational science and educational research that preserves the analysis of meaning and subjectivation while overcoming traditional scalability constraints.