Enhanced speaker-turn aware hierarchical model for automated classroom dialogue act classification

Dialogue act classification (DAC) is a pivotal task in natural language processing, providing deep insights into communication patterns. In classroom settings, analyzing dialogues and delivering timely feedback play a crucial role in improving teaching practices and fostering effective educational interactions. Traditional manual classroom observation is both costly and prone to bias, underscoring the need for automated dialogue analysis. However, recent automated classroom DAC often neglect the importance of speaker turns. To address this gap, we propose the Speaker-Turn Aware Hierarchical Model (SA-DAC), which leverages speaker embeddings to enrich contextual understanding and improve classification accuracy. We further introduce a new benchmark dataset, IRE-9, designed around the initiate-response-evaluate (IRE) encoding framework, to advance research in dialogue analysis. Comprehensive experiments demonstrate that SA-DAC surpasses state-of-the-art methods, delivering superior accuracy and scalability. Moreover, we develop an automated classroom analysis system powered by SA-DAC, which generates detailed, multi-dimensional reports to assist educators in improving their teaching strategies.