Cognitive Diagnosis Driven Programming Ability Evaluation

With the widespread adoption of 5G and IoT, computers are extensively used across various societal domains, bolstering programming education. Numerous programming novices complete introductory courses through online platforms, generating copious programming-related data. Leveraging these data for more precise student programming ability assessments presents a new industry challenge. Cognitive diagnosis stands as a mature evaluation model in this context, enabling students to gauge their knowledge based on their answer records. Nonetheless, deficiencies persist in current programming ability evaluation algorithms. Firstly, deep learning-based algorithms lack interpretability, which is crucial for guiding students’ learning in the education field. Secondly, the unstructured nature of massive programming platform data necessitates feature screening for cognitive diagnosis models. Finally, the relationship structure between programming knowledge points is often overlooked, resulting in inadequate exploration of the interplay between topics and knowledge points. In order to solve the above problems, we focus on the programming ability evaluation algorithm based on cognitive diagnosis. To solve the problem of low interpretability of the neural network model alone, we introduce the definitions of knowledge points and student ability factors in the cognitive diagnosis model to assist modeling. A cognitive diagnosis model combined with knowledge structure reasoning is proposed to model the structure of knowledge points. As an embedded layer, it is combined with the process of neural network training to finally obtain the mastery of students’ knowledge points and understand the structural relationship between each knowledge point. Compared to the state-of-the-art NCD algorithm, our solution enhances accuracy and consistency by 3.5% and 7.2%, respectively.