An intelligent tutoring system using Bayesian Networks and Content-based recommendation

Abstract

This dissertation looks to implement the Artificial Intelligence technique of uncertainty management and probabilistic reasoning using Bayesian Networks to develop an Intelligent Tutoring System. This ITS is able to learn the student's learning abilities and adapt its teaching techniques accordingly, to better suit the educational process of the student. The main aim is to provide students with an individualised tutoring solution by emulating a private tutor, and teaching the student using the best learning techniques. This solution is also extended to encompass Content-Based Recommendation. This aspect of the ITS is aimed at providing a tool for students to better manage and harness the potential of the unlimited number of online resources available. The approach used to implement the C-B Recommendation was by using the TF-IDF vector space. The testing and evaluation results have successfully supported the ideas behind this dissertation, as well as the actual implementation prototype. A number of testing methods were implemented to ensure that the entire system and its representations are properly evaluated. The results showed a roughly 90% student improvement when using the ITS as opposed to using traditional techniques. About 85% of the students who were used for the testing spoke in favour of the ITS as well as the Recommendation tool. With regards to the Bayesian Network implementation, this dissertation has contributed the following: the use of probabilistic reasoning and uncertainty management is a useful tool in order to model a private tutor and develop an Intelligent Tutoring System capable of learning student styles and adapting. It has also contributed a novel way of combining the Network modeling the Knowledge Domain and the Network modeling the Pedagogical Techniques together, resulting in one Bayesian Network modeling the whole ITS domain. Another contribution is the object-oriented approach, as the ITS was modeled over a shell of the Knowledge Domain and the Bayesian Network.