國家教育研究院

　　本研究是一年期計畫，探討非監督式認知診斷模式估計演算法則，開發監督式與半監督式認知診斷模式估計演算法，透過模擬與實證資料驗證本研究所開發監督式、半監督式認知診斷模式估計演算法之成效。目前的認知診斷模式均是屬於非監督式學習，本研究應用監督式與半監督式學習之方法，在認知診斷模式估計過程中，加入已標記之訊息，探討其估計成效。研究結果顯示：在認知屬性的估計方面，以監督式的演算法則估計誤差最小，非監督式的演算法則估計誤差最大，半監督式則介於兩者之間。當固定題數、認知屬性數與樣本人數時，隨著訓練樣本人數的增加，試題參數的估計誤差也隨著降低，認知屬性之估計而亦有類似的結果。當固定題數、認知屬性數與訓練樣本人數時，樣本人數2000人與樣本人數1000人，兩者之估計成效差異不大。增加題數有助於增加認知屬性與試題參數估計之準確度。

　　This study is proposed for one-year period of research implementation. The project focused on cognitive diagnostic models estimation with unsupervised, supervised and semi-supervised learning. This study developed two new algorithms, supervised and semi-supervised estimations, based on cognitive diagnostic models. A simulation study and a real data are used to evaluate the performance of the proposed algorithms. The results showed that the performance of supervised learning preformed better than that of semi-supervised and unsupervised estimations in terms of estimating attributes. Under condition of fixing the number of items, attributes, and subjects, while the number of training sample increased, the root mean square errors of items and attributes decreased. The performance of estimating attributes is not affect by the sample size in this simulation study (N=1000 and N= 2000); however, increasing the number of items can improve the performance of estimating attributes and item parameters.