Collaborative learning is based on groups of students working together with traditional and computer-based tools or applications. We have found that to make these supporting applications more effective we need to address the problem of automating group awareness in CSCL applications by estimating group arrangements from location sensors and the history of interaction. This contextual information can enable the construction of applications that facilitate communication among group members in synchronous and collocated collaborative learning activities. We used data traces collected from the study of students‟ behavior to train and test an intelligent system. Results show that context-information can be effectively used as a basis for a middleware for automating group management. Inferring group membership is technically feasible, can be integrated in group-support applications and can be used in real-world settings.