Neural cognitive diagnosis modeling incorporating response times

Cognitive diagnosis is a fundamental issue in the field of intelligent education, aiming to identify students' mastery of specific knowledge concepts. With computerized testing, response time (RT) is a process data that can be collected. Incorporating RT in cognitive diagnosis assessment can enhance diagnostic accuracy. However, RT is only considered in traditional statistical cognitive diagnosis models. Compared with traditional statistical diagnostic models, cognitive diagnosis models based on neural networks have advantages such as high precision and strong generalization ability. Therefore, this paper proposes a JRT-NCD (joint response times neural cognitive diagnosis) model that uses neural networks to model the complex nonlinear interactions between exercises and students and incorporates RT as a new feature to refine diagnostic results on student abilities. Research findings on three datasets of PISA2012, 2MFC, and ASSIST09 indicate that: (1) In comparison with traditional statistical models, neural networks have better fitting capabilities for real complex nonlinear data; (2) compared to the NCD model that disregards RT, JRT-NCD achieves higher diagnostic accuracy while maintaining its interpretability, and reduces the misleading effects of "overspeed behavior" on the diagnostic results.