Using natural language processing to analyze elementary teachers’ mathematical pedagogical content knowledge in online community of practice

Abstract

This study focuses on how teachers’ pedagogical content knowledge (PCK) of mathematics may differ depending on teacher interactions in an online teacher community of practice (CoP). The study utilizes data from 26,857 posts collected from the South Korean self-generated online teacher CoP, ‘Indischool’. This data was then analyzed using natural language processing techniques; specifically, text classification with word2vec, BERT, and machine learning classifiers was used. The results indicate that the texts of posts can predict the level of teacher interactions in the online CoP. BERT embedding and classifier exhibited the best performance, ultimately achieving an F1 score of .756. Moreover, topic modeling utilizing BERT embedding is used to uncover the specific PCK of teachers through high- and low-interaction posts. The results reveal that high-interaction posts with numerous likes and replies demonstrate more in-depth reflections on teaching mathematics and refined PCK. This study makes two significant contributions. First, it applies a data science framework that allows for the analysis of real data from an actual online teacher community. Secondly, it sheds light on the intricacies of knowledge management in an online teacher CoP, an area that has to this point received limited empirical attention.