A1416
Title: Q-matrix estimation in cognitive diagnostic models by using overlapping clustering
Authors: Atsuro Kimoto - Doshisha University (Japan) [presenting]
Jun Tsuchida - Kyoto Womens University (Japan)
Hiroshi Yadohisa - Doshisha University (Japan)
Abstract: Cognitive diagnostic models (CDMs) are educational measurement models that assess individuals' mastery of certain cognitive skills (hereafter, attributes) from their responses. A key objective of CDMs is to extract the attribute profile that represents each individual's attribute mastery status. To extract the attribute profile, the Q-matrix, which associates each test item with the attributes, should be developed before applying the CDMs. While domain experts generally develop the Q-matrix, it can be mis-specified, potentially leading to incorrect interpretations. Various data-driven Q-matrix estimation methods have been proposed to address this issue. However, estimating the Q-matrix appropriately is difficult because the elements of the Q-matrix are binary variables, and multiple attributes can be associated with a single test item. To address this difficulty, a novel Q-matrix estimation method is proposed based on the elements of the Q-matrix as a membership indicator of overlapping clustering. Using the overlapping clustering algorithm is expected to improve both accuracy and computational efficiency. Moreover, the proposed method can be applied to several types of CDMs. The results of numerical experiments show that the proposed method is superior in accuracy and computational efficiency compared to existing methods.
