CLARE: a computer-supported collaborative learning based on the thematic structure of scientific text

This dissertation presents a computer-based collaborative learning environment, called CLARE, that is based on the theory of learning as collaborative knowledge building. It addresses the question, "what can a computer do for a group of learners beyond helping them share information?" CLARE differs from virtual classrooms and hypermedia systems in three ways. First, CLARE is grounded on the theory of meaningful learning, which focuses the role of meta-knowledge in human learning. Instead of merely allowing learners to share information, CLARE provides an explicit meta-cognitive framework, called RESRA, to help learners interpret information and build knowledge. Second, CLARE defines a new group process, called SECAI, that guides learners to systematically analyze, relate, and discuss scientific text through a set of structured steps: summarization, evaluation, comparison, argumentation, and integration. Third, CLARE provides a fine-grained, non-obtrusive instrumentation mechanism that keeps track of the usage process of its users. Such data forms an important source of feedback for enhancing the system and a basis for rigorously studying collaboration learning behaviors of CLARE users. CLARE was evaluated through sixteen usage sessions involving six groups of students from two classes. The experiments consist of a total of about 300 hours of usage and over 80,000 timestamps. The survey shows that about 70% of learners think that CLARE provides a novel way of understanding scientific text, and about 80% of learners think that CLARE provides a novel way of understanding their peers' perspectives. The analysis of the CLARE database and the process data also reveals that learners differ greatly in their interpretations of RESRA, strategies for comprehending the online text, and understanding of the selected artifact. It is also found that, despite the large amount of time spent on summarization (up to 66%), these learners often fail to correctly represent important features of scientific text and the relationships between those features. Implications of these findings at the design, empirical, and pedagogical levels are discussed.