An Approach to Generating Adaptive Feedback for Online Formative Assessment.

In this paper, we propose a novel approach to generating adaptive feedback by identifying the chain of weakest learning objectives to a learner working in a domain. It combines the domain model based expert-driven model with question-answering based data-driven model. The domain model is an AND/OR graph of domain knowledge structure based on the revised Bloom’s taxonomy, defining the learning objectives of the domain and the corresponding pre-requisite relationships. The adaptive formative assessment process uses an improved Top-Two Thompson sampling algorithm for solving the best arm identification problem in the multi-armed bandit framework. The simulation results show the feasibility and performance of the proposed approach.