Assessing and Scaffolding Collaborative Learning in Online Discussions

In this paper we present two computational approaches that can be used characterize and measure online threaded discussions and demonstrate that they can objectively validate student-reported differences in collaborative learning between tutor-scaffolded and non-scaffolded discussion activities. The first approach, thread profiling, is used to characterize user interactions that tend to broaden and deepen discussions, and gives insight into how tutors participate in discussions. The second approach, which uses a natural language discourse processor, is used to compare the rhetoric of tutors and students, and shows that tutors consistently use more attributions, elaborations, and enablements to scaffold discussions. To test these ideas we processed twenty-four online activities, constituting over one thousand message posts, during a course at the British Open University. These computational methods and findings have application in virtual tutoring systems and the automated assessment of discussions.