A Method to Reduce Semantic Redundancy of Question in Bee Algorithm-based Test Assembly Task

Abstract

                Redundancy of questions in a test-form generated from an automated test assembly is a problem leading to unfairness in an examination. This paper thus proposes a method to enhance automated test assembly using semantic similarity calculation for reducing a probability in assembling similar questions in the same test-form. The semantic similarity is designed to use as an objective function in Bee algorithm which reliably performs combinatorial optimization to manage computation complexity and required constraints in test assembling. The experiments were conducted with 400 questions in information technology domain in comparison of using and not using the semantic similarity. The results showed that applying semantic similarity calculation in a test assembly could greatly reduce a number of co-occurred similar questions for more than 90% in average. In addition, it does not increase computational times, and it can produce test-forms meeting specified constraints such as an equivalently distributed test length.