Background The global growth in Chinese language learners presents systematic challenges to Chinese classroom interaction evaluation. Key limitations include inadequate accuracy in speech recognition and shallow interaction analysis. Methods This study proposes a novel framework that integrates three components: the Whisper-large-v3 model for automated speech recognition, PyAnnote for speaker diarization, and social network analysis for assessing teacher-student verbal interactions. We developed an evaluation model based on directed graph incorporating two core dimensions: interaction breadth and depth. Interaction breadth is quantified by the network density parameter, while interaction depth is measured using interaction frequency and duration, which are defined as edge weights between graph nodes (representing teachers and students). Results The framework was pilot-tested on 24 WAV-format classroom dialog recordings. These recordings were transcribed from publicly available videos of China’s Teaching Ability Contest in the year of 2020 on Bilibili, with all personally identifiable information removed. Results reveal that teachers emerge as the core figures in classroom interactions. Denser directed graphs correlated with more frequent, longer interactions and higher verbal interaction scores. This work contributes a scalable, Artifical Intelligence (AI)-driven dynamic assessment framework to support teachers in enhancing their teaching effectiveness.
Zhang et al. (Thu,) studied this question.