Data-driven teaching assessment in inquiry-based learning by topic modeling

Abstract

Inquiry-based learning (IBL), wherein learning is driven by a student’s inquiry, becomes popular in high school in many developed countries. IBL environments are also considered as a promising field of learning analytics or educational data mining. Until today, many researchers tried various machine learning methods to the process of IBL. However, student IBL outcomes written as a text have rarely been analyzed quantitatively. This paper aims to quantitatively assess the teaching of IBL via unsupervised machine learning method with natural language processing. Here, we propose a novel method for teaching assessment with topic modeling. Since educational assessment needs two kinds of information, i.e., what teachers want students to learn and how students change, we also plan to calculate the correlation between the topic and teacher’s rating and the correlation between the topic and the posted year of the documents. In a preliminary experiment, we collected students’ graduate theses over 31 years in a high school (N=3,328), we confirmed the tendency that a topic evaluated by teachers was also a topic which is popular among students. We think that the method of using the topic model applied to a large collection of output texts would be a good way to assess the teaching of IBL.