Towards Predictive Modelling of Student Affect from Web-Based Interactions
Manolis Mavrikis, Antony Maciocia, John R. Lee · UCL Discovery (University College London) · 2007
This paper presents the methodology and results of a study conducted in order to establish ways of predicting students' emotional and motivational states while they are working with Interactive Learning Environments (ILEs). The interactions of a group of students using, under realistic circumstances, an ILE were recorded and replayed to them during post-task walkthroughs. With the help of machine learning we determine patterns that contribute to the overall task of diagnosing learners' affective states based on observable student-system interactions. Apart from the specific rules brought forward, we present our work as a general method of deriving predictive rules or, when there is not enough evidence, generate at least hypotheses that can guide further research.