Assessment intelligence in small group learning

Assessment of groups in CSCL context is a challenging task fraught with many confounding factors collected and measured. Previous documented studies are by and large summative in nature and some process-oriented methods require time-intensive coding of qualitative data. This study attempts to resolve these problems for teachers to assess groups and give timely feedback. We first operationalize activity theory to holistically frame group work by breaking it down into six dimensions. The captured log data by the collaborative software is mapped with these dimensions and as a result, six measures are generated with a semantic background. Next, we employ a relatively new clustering algorithm - spectral clustering - to categorize groups with similar behaviors, which not only allows us to consider the six indicators simultaneously but also has the capability to deal with large online context. The spectral clustering result is compared with traditional algorithms and demonstrates better assessment accuracy. Furthermore, since the whole process is automated and the group performance indicators are grounded in a meaningful backdrop, it enables teachers to offer concrete and personalized help in a real-time format. Theoretical and practical implications are then discussed.