A Bandits Approach to Intelligent Tutoring Systems using Concept Evolution Estimation

Sudha S, Arun Rajkumar

Abstract: With the huge number of learning resources available online today, the Intelligent Tutoring Systems (ITS) are of great need more than ever. An ITS is a system that personalizes the course contents to each learner. In this paper, we address the problem of suggesting an effective & efficient learning sequences to learners based on their knowledge levels. We take a multi-armed bandits approach to action choosing where we suggest that action which has the highest estimated learning outcome at each step. We model the actions as Beta distributions & the learners’ knowledge level as concept vectors. We also learn the prerequisite relationships that can exist among the concepts automatically. We propose a novel algorithm that achieves the goal efficiently. Our experimental results show that our algorithm’s performance is comparable to that of the optimal algorithm.