Lessons clustering using topics inferred by unsupervised modeling from textbooks

Abstract

Analyzing the content and quality of teacher and students’ talk has been an active area of educational research. In this context, the importance of temporal analysis of teaching learning events has been growing up. Previous work has proposed a method that automatically describes teacher’s talk using an unsupervised machine learning model to infer topics from school textbooks. To describe teacher talk, the machine learning method measures the appearance of the inferred topics throughout each lesson. We propose a clustering method based on a modification of the method described above. The modification consists in considering super topics (Content, Administration/Feedback, Other), which will describe teacher talk more generally. Then, we cluster using ‘K-means’ with the Dynamic Time Warping metric since the lessons are dynamic phenomena that occur over time. Finally, we propose a way to visualize the center of the clusters to analyze them. We apply the proposed method to a collection of natural science lesson transcriptions, and we analyze and discuss the clusters obtained.