EFG - 13

Analyzing effective teaching and learning in real-life situations like school classrooms is complex. Addressing learners' heterogeneity and assessing teaching effectiveness are important challenges in both practice and research. Approaches such as adaptive teaching (e.g., continuous diagnosing, monitoring, and supporting learners individually; Corno, 2008; Plass & Pawar, 2020) or the integration of different assessment methods like student ratings or observation data have been discussed for decades and yet remain exceptions in educational practice. In contrast to existing initiatives strongly focusing on learning analytics or specific aspects such as collaborative learning (special issue; EFG 12), we will use a broad range of existing ML algorithms to assess teaching effectiveness (e.g., facial and speech emotion recognition) of existing multimodal data (e.g., video, transcripts, eye-tracking) to get a cost- and time-efficient but holistic understanding of complex teaching and learning situations (Fütterer et al., 2023; Hou et al., 2024). The results can support learning (e.g., immediate feedback: Fütterer et al., 2024). The main goal of this EFG is to explore the potential of new technological approaches to automatically assess teaching effectiveness using multimodal data in interdisciplinary research to master urgent educational challenges like personalized learning. We bring together researchers from the fields of educational psychology and computer science/ML to (1) identify available multimodal data, systematically compile analysis strategies (e.g., gather pre-trained models), and research desiderata in a conceptual paper, (2) identify approaches (e.g., combining algorithms) to reliably and validly predict important educational constructs (e.g., cognitive engagement, mind wandering), (3) identify innovative approaches to answer research questions not possible to answer without multimodal data (e.g., interaction of cognitive, emotional, behavioral components of engagement and their relation to teaching effectiveness), and (4) test the effectiveness of ML systems in experiments using information from multimodal data (e.g., real-time feedback on teaching effectiveness for teachers in dashboard systems).