Beyond Tutor Logs: Exploring Multiple Streams of Data to Identify Self-Regulated Learning

• In Event: Self-Regulated Learning Analytics: Aligning Data and Their Treatment to the Assumptions of Theory

Abstract

Traditionally, models of self-regulated learning have been derived from data logs, but the predictive power of these models is limited to the actions that the online systems can log. In many instances, the online systems are used in a classroom context and computer-collected data cannot capture all important learning phenomena. A student working at a computer might be working independently with few outside influences. Alternatively, the student might be in a lively classroom, with other students around her, talking and even offering suggestions. Between sessions at a computer, they might encounter other relevant experiences. The student might participate in teacher-led classroom instruction, work in a small group with other students, or listen to teacher lectures. Data that capture these other experiences would add to and complement computer-collected data. In some cases, these additional data may lead to absolutely critical insights. Most educational data mining analyses omit essential contextual data for a number of reasons. Data from different sources are difficult to integrate, data about classroom activities are challenging to collect and code, and data at different grain sizes and timescales are difficult to integrate. We are working to leverage and extend techniques of educational data mining across multiple modalities of data (computer log files, video, and written artifacts), contexts (online systems, peer-to-peer interactions, classroom activities) and timescales (milliseconds, minutes, class periods, units) in order to develop more robust and predictive models of student learning and behavior.
In this work we report on our progress of this NSF funded project called “Learning Linkages: Integrating data streams of multiple modalities and timescales” as it relates to self-regulated learning. We are creating a more complete picture of the limits of existing models in the context of self-regulated learning by looking at short term and long term learning outcomes. As a result, these new models should be better suited to generate actionable knowledge for systems, students, teachers, and researchers.

Authors

• John Stamper, Carnegie Mellon University

• Ran Lui, Carnegie Mellon University

• Jodi Davenport, WestEd

• Bruce Sherin, Northwestern University

• Danielle S. McNamara, Arizona State University