Modelling and identifying collaborative situations in a collocated multi-display groupware setting


Roberto Martinez Maldonado, James Wallace, Judy Kay and Kalina Yacef

Paper type: 
Full paper
2. 11:00-12:30, Wednesday 29 June


Detecting the presence or absence of collaboration during group work is important for providing help and feedback during sessions. We propose an approach which automatically distinguishes between the times when a co-located group of learners, using a problem solving computer-based environment, is engaged in collaborative, non-collaborative or somewhat collaborative behaviour. We exploit the available data, audio and application log traces, to automatically infer useful aspects of the group collaboration and propose a set of features to code them. We then use a set of classifiers and evaluate whether their results accurately match the observations made on video-recordings. Results show up to 69.4% accuracy (depending on the classifier) and that the error rate for extreme misclassification (e.g. when a collaborative episode is classified as non-collaborative, or vice-versa) is less than 7.6%. We argue that this technique can be used to show the teacher and the learners an overview of the extent of their collaboration so they can become aware of it.


Data Mining, Group Modelling, Collaborative Learning