Decision Support Systems Laboratory - Department of Production Engineering and Management (GREECE)
About this paper:
Appears in: ICERI2010 Proceedings
Publication year: 2010
Pages: 5214-5222
ISBN: 978-84-614-2439-9
ISSN: 2340-1095
Conference name: 3rd International Conference of Education, Research and Innovation
Dates: 15-17 November, 2010
Location: Madrid, Spain
Learning objectives have been researched in cognitive psychology and in education, in order to specify the goals of a learning process. Bloom’s taxonomy provides a description of learning objectives in 6 stages of scaling complexity. Bayesian networks and decision theory have been used extensively in Intelligent Tutoring Systems in order to model uncertainty on the one hand and learner or tutor preferences on the other, while diagnosing the learners’ cognitive state. In this paper we evaluate a model for task-based diagnosis of cognitive achievement, based on learning objectives, static Bayesian networks, decision theory and multi-attribute utility theory. The cognitive achievement of learning objectives is tested against observed learner behaviour in a non-deterministic manner, due to the inherent uncertainty in the problem. The proposed model is domain-independent since it assumes that learning material consists of generic Learning Objects associated with specific learning objectives. Multi-attribute utility theory is used for combining learning evidence with tutor preferences, thus providing for different tutors and different types of tutoring. Bayesian networks are used for handling the uncertainty of the diagnosis by calculating the posterior probability of achieving a learning objective based on the available evidence. Decision theory is used for expressing the pedagogical utility of available actions. The evaluation parameters include preferences regarding the value of evidenced learner behaviour such as assessment results, affective states, the learner’s own belief and the time-on-task, as well as the pedagogical utility of available actions. All learning evidence is modelled as discrete-state variables. Assessment results are modelled as binary or multiple-state variables. Affective states range from negative to positive, while the time-on-task is discretized in a scale from ‘too little time’ to ‘a lot of time’. The learner’s own belief about her cognitive state is included in the model, as an important feature of an open learning process. The learner’s belief is expressed through his post-task activity with respect to the learning goal at hand. The model is used for the selection of suitable remediation Learning Objects (LO’s) or for moving on to the next learning goal, in an intelligent learning environment. Data from real tutoring sessions in introductory programming courses and real assessment results are used in order to test the validity of the model.
Intelligent learning, cognitive diagnosis, Bayesian networks, decision theory, multi-attribute utility theory.