Adaptive group formation to promote desired behaviours

Background: There is substantial literature that shows the benefits of collaborative work, though these benefits vary enormously with circumstances. Irrespective of their structure and composition, groups usually exist for a particular reason and implicitly or explicitly target one or more outcomes. The achievements of group outcomes depend on many factors, including the individual behaviour of each group member. These behaviours are, in turn, affected by the individual characteristics, the context and the group composition. Constructing groups in a way that maximises the achievement of a specific outcome is complex with the optimal group composition depending on the attributes of the group members. Previous work has in most cases considered group formation based on one particular attribute, such as learning style, gender, personality, etc. Less common are instances of group formation rules being adjusted systematically to accommodate changes in an individual's attributes or disposition. Purpose: This paper considers how the multi-factorial nature of group performance and the variations in desired behaviour across different circumstances can be addressed within a consistent framework. Design/Method: The methodology consisted of two main stages. In the first stage, a simulation was encoded in MatLab to assess the conceptual approach of progressively updating rules for group formation. The method uses an unsupervised learning algorithm and correlation factors between quantifiable group characteristics (average age, degree of motivation, etc.) and resultant behaviours of the groups that are actually formed (level of dialogue, interface interactions, etc.) to update the rules used for group formation, and hence progressively construct groups that are more likely to behave in desired ways. The second stage involved an evaluation of this approach in a real world scenario using remotely accessible laboratories where engineering students voluntarily participated in a study in April 2012. Results: The simulation results show that under certain conditions the desired behaviour chosen with the intention of improving specific learning outcomes can be optimized and that groups can be constructed that are more likely to exhibit desired behaviour. The paper also reports preliminary evidence that shows the feasibility of this approach in selecting group participants in an engineering class to promote a desired outcome in this case independent learning. Conclusions: This study demonstrates the feasibility of using a set of individual characteristics of group members to form groups that are more likely to have desired group behaviours and that these characteristics can be monitored and updated to dynamically alter group formation to account for changes in any individual's characteristics. This has potential to allow groups formation decisions to be made dynamically to achieve a desired outcome, for example promote collaborative learning.