Latham, A.M. (2011) Personalising Learning with Dynamic Prediction and Adaptation to Learning Styles in a Conversational Intelligent Tutoring System. Ph.D. Manchester Metropolitan University.
This thesis presents research that combines the benefits of intelligent tutoring systems (ITS), conversational agents (CA) and learning styles theory by constructing a novel conversational intelligent tutoring system (CITS) called Oscar. Oscar CITS aims to imitate a human tutor by implicitly predicting individuals’ learning style preferences and adapting its tutoring style to suit them during a tutoring conversation.
ITS are computerised learning systems that intelligently personalise tutoring based on learner characteristics such as existing knowledge and learning style. ITS are traditionally student-led, hyperlink-based learning systems that adapt the presentation of learning resources by reordering or hiding links. Research suggests that students learn more effectively when instruction matches their learning style, which is typically modelled explicitly using questionnaires or implicitly based on behaviour. Learning is a social process and natural language interfaces to ITS, such as CAs, allow students to construct knowledge through discussion. Existing CITS adapt tutoring according to student knowledge, emotions and mood, however no CITS adapts to learning styles.
Oscar CITS models a human tutor by directing a tutoring conversation and automatically detecting and adapting to an individual’s learning styles. Original methodologies and architectures were developed for constructing an Oscar Predictive CITS and an Oscar Adaptive CITS. Oscar Predictive CITS uses knowledge captured from a learning styles model to dynamically predict learning styles from an individual’s tutoring dialogue. Oscar Adaptive CITS applies a novel adaptation algorithm to select the best tutoring style for each tutorial question. The Oscar CITS methodologies and architectures are independent of the learning styles model and subject domain. Empirical studies involving real students have validated the prediction and adaptation of learning styles in a real-world teaching/learning environment. The results show that learning styles can be successfully predicted from a natural language tutoring dialogue, and that adapting the tutoring style significantly improves learning performance.