Please use this identifier to cite or link to this item:
|Title:||High Accuracy Detection of Collaboration From Log Data and Superficial Speech Features|
|Authors:||Viswanathan, Sree Aurovindh|
|Publisher:||Philadelphia, PA: International Society of the Learning Sciences.|
|Citation:||Viswanathan, S. A. & Vanlehn, K. (2017). High Accuracy Detection of Collaboration From Log Data and Superficial Speech Features In Smith, B. K., Borge, M., Mercier, E., and Lim, K. Y. (Eds.). (2017). Making a Difference: Prioritizing Equity and Access in CSCL, 12th International Conference on Computer Supported Collaborative Learning (CSCL) 2017, Volume 1. Philadelphia, PA: International Society of the Learning Sciences.|
|Abstract:||Effective collaborative behavior between students is neither spontaneous nor continuous. A system that can measure collaboration in real-time may be useful. For instance, it could alert an instructor that a group needs attention. We tested whether superficial measures of speech and user interactions of students would suffice for measuring collaboration. As pairs of students solved complex math problems on tablets, their speech and tablet gestures were recorded. These data and multi-camera videos were used by humans to code episodes as collaborative vs. various kinds of non-collaboration. Using just the speech and tablet log data, several detectors were machine learned. The best had an overall accuracy of 96% (Kappa=0.92), which is higher than earlier attempts to use speech and log data for detecting collaboration. The improved accuracy appears to be due both to our analytic methods and to the particular mathematical task, which involves moving objects.|
|Appears in Collections:||CSCL 2017|
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.