Bayesian expert systems and multi-agent modeling for learner-centric Web-based education

Abstract

Online distance education provides students with a wealth of information. When students submit course-related search term queries, the search engine returns the search hits based on keyword and topic match. A student's particular learning style is not taken into consideration. For instance, a visually oriented student may benefit more than others from viewing videos and interacting with simulations. We address this problem by designing and developing a knowledge-based system for the initial assessment of students' learning styles. Each student's membership in a learning style dimension (e.g. visual or verbal) is estimated probabilistically. We reach this probability value by using a sequential Bayesian approach to administer a dynamic questionnaire that aims to attain a desired confidence level estimate with the minimal number of questions. A multi-agent online tutoring system uses this initial learning style model to start suggesting learning material matching the student's style. Each agent is an expert in a learning style dimension and can suggest the learning materials matching the student's style. In addition, these agents closely follow the student's evolving preferences and continuously update the stochastic model based on the student's online activities. When the student searches for course material, the multi-agent system delivers the search matches in a cycle-free preference order influenced by the students' multi-dimensional learning style model.