Analysing Frequent Sequential Patterns of Collaborative Learning Activity Around an Interactive Tabletop. Nominee for Best Paper Award

Electronic traces of activity have the potential to be an invaluable source to understand the strategies followed by groups of learners working collaboratively around a tabletop. However, in tabletop and other co-located learning settings, high amounts of unconstrained actions can be performed by different students simultaneously. This paper introduces a data mining approach that exploits the log traces of a problem-solving tabletop application to extract patterns of activity in order to shed light on the strategies followed by groups of learners. The objective of the data mining task is to discover which frequent sequences of actions differentiate high achieving from low achieving groups. An important challenge is to interpret the raw log traces, taking the user identification into account, and pre-process this data to make it suitable for mining and discovering meaningful patterns of interaction. We explore two methods for mining sequential patterns. We compare these two methods by evaluating the information that they each discover about the strategies followed by the high and low achieving groups. Our key contributions include the design of an approach to find frequent sequential patterns from multiuser co-located settings, the evaluation of the two methods, and the analysis of the results obtained from the sequential pattern mining.