[1M4-OS-14a] OS-14

[1M4-OS-14a-04] Automated parallel test forms assembly for time limitation using Deep-IRT with prediction of item response time

〇Fumiya Ishikawa¹, Kazuma Fuchimoto¹, Wakaba Kishida¹, Emiko Tsutsumi², Maomi Ueno¹ (1. The University of Electro-Communications, 2. The University of Tokyo)

Keywords:Computer Based Testing, Deep Learning

Recently, through advances in the study of artificial intelligence, automated assembly of parallel test forms, for which each test has an equivalent measurement accuracy of the examinee's ability but with a different set of items, have been put to practical use. An important issue for automated assembly is to assemble as many tests as possible. To solve this issue, the maximum clique algorithm using the integer programming method is known to be able to assemble the greatest number of tests with the highest measurement accuracy of the examinee’s ability. However, this automated assembly does not control a time limitation in tests. As a result, each test decreases the measurement accuracy because the examinee can not finish the test within the time limitation. To address this problem, we propose a new automated parallel test forms assembly for the time limitation using Deep-IRT with prediction of item response time. The main idea of the proposed method is to employ the prediction of item response time for the examinee using Deep-IRT to constraint in parallel test forms. Numerical experiments demonstrate that up to 95% of examinees finish the examination within the time limitation.