Preference Cognitive Diagnosis for Predicting Examinee Performance

Abstract

Cognitive diagnosis can discover the examinees' knowledge state for predicting their performance (i.e., scores). However, cognitive modeling only considers the examinees' response logs on exercises, so the prediction accuracy cannot be guaranteed. Actually, examinees usually read some text learning materials before responding exercises to enhance their performance. These specific learning materials will reveal examinees' learning preferences, which will also reflect their knowledge state to some extent. To this end, we propose a Preference Cognitive Diagnosis (Preference CD) method, which combines learning preferences with cognitive diagnosis to evaluate examinees' knowledge state so as to gain accurate prediction results. Specifically, first, we utilize the cognitive diagnosis model to obtain examinees' mastery for exercises. Then, we establish the keyword vector model that can extract examinees' learning preferences information from the contents of learning material read by examinees. Further, we quantify the extracted examinees' learning preferences information as their preference degree for exercise. Finally, we project the generation of exercise scores by combining examinees' mastery and preference degree for exercises. Experimental results on learning behavior data from the actual operating system show PreferenceCD can predict examinees' performance more accurately.