Graph convolutional network for automatic detection of teachers’ nonverbal behavior

Accurately discerning teachers' nonverbal behaviors is essential to facilitate their reflective practice and professional growth. Yet, conventional manual approaches to observe and assess these behaviors are exceedingly time-consuming and resource-intensive. Furthermore, limited research has examined the precise and automated detection of teachers' nonverbal behaviors in classroom videos. This problem is exacerbated by the absence of a universally accepted framework for categorizing such behaviors, as well as a dearth of standardized datasets and corresponding algorithms. To address these challenges, this study investigates methods for the automatic detection of teachers' nonverbal behaviors. First, we theoretically define five common categories encompassing teachers' nonverbal behaviors, furnishing a structured framework for their analysis. Subsequently, in our methodology, we focus on primary and secondary school teachers, utilizing classroom video recordings as our primary data source. This study meticulously gathers and annotates datasets comprising teachers' nonverbal behaviors and devises an innovative method grounded in graph convolutional networks for their automated detection. To evaluate the effectiveness of our approach, we compare it with other state-of-the-art methods, highlighting its advantages. In practice, our method holds significant potential for real-world applications within authentic teaching contexts, offering the capability to automatically detect teachers' behaviors. These automated insights can serve as a valuable tool to guide and refine teaching practices, fostering more productive teaching outcomes.