Determining quality and distribution of ideas in online classroom talk using learning analytics and machine learning

Abstract

The understanding of online classroom talk is a challenge even with current technological advancements. To determine the quality of ideas in classroom talk for individual and groups of students, a new approach such as precision education will be needed to integrate learning analytics and machine learning techniques to improve the quality of teaching and cater interventive practices for individuals based on best available evidence. This paper presents a study of 20 secondary school students engaged in asynchronous online discourse over a period of two weeks. The online discourse was recorded and classroom talk was coded before undergoing social network analysis and k-means clustering to identify three types of ideas (promising, potential, and trivial). The quality and distribution of ideas were then mapped to the different kinds of talk that were coded from the online discourse. Idea Progress Reports were designed and trialed to present collective and individual student’s idea trajectories during discourse. Findings show that the majority of ideas in exploratory talk are promising to the students, while ideas in cumulative and disputational talks are less promising or trivial. Feedback on the design of the Idea Progress Reports was collected with suggestions for it to be more informative and insightful for individual student. Overall, this research has shown that classroom talk can be associated with the quality of ideas using a quantitative approach and teachers can be adequately informed about collective and individual ideas in classroom talks to provide timely interventions.