Detection of Student Modelling Anomalies

Abstract

As the modern TEL tools gain wider adoption in real educational contexts, they start facing important practical problems. One such problem for adaptive educational systems is the reliability of their student modelling mechanisms. Even when such a mechanism has been tested and calibrated to represent students’ knowledge reasonably well, the student herself can become a source of problems. Students can use the system in a non-intended way, exhibit long periods of off task behaviour, try gaming the system, seek help of parents or peers, etc. Such usage patterns will manifest themselves in sequences of activity that do not represent student abilities and will result in student modelling anomalies causing subsequent suboptimal adaptive interventions from the system. This would be very important for a system that is used in real classrooms with younger children, especially, when it is also available at home as a supporting tool for independent work. This paper reports a study of such a system – Math Garden. Several user modelling anomalies have been detected in its logs. First steps towards building an automated tool for on-the-fly student modelling anomaly detection are reported.