Using MPE with Bayesian Network for Sub-optimization to Entropy-Based Methodology

Many researchers show that the properties of Bayesian network and information theory such as entropy between dichotomous concepts and test items generalize some common intuitions about item comparison, and provide principled foundational to design item-selection heuristics for adaptive testing in computer-assisted educational systems. But entropy-based heuristic methodology could be too time-consuming as interesting variables with high dimensions to perform in practical situations. Hence the goal of this paper is trying to modify entropy-based heuristic methodology as a new form using most probable explanation (MPE) with Bayesian network to overcome this problem and to hold considerably performance for constructing decision items tree for adaptive testing in computer-assisted educational systems. Experiment results show that the proposed new methodology, named MPE-entropy-based heuristic methodology, can reduce the time-complexity and lose little performance.