Fully Adaptive Framework: Neural Computerized Adaptive Testing for Online Education

Computerized Adaptive Testing (CAT) refers to an efficient and personalized test mode in online education, aiming accurately measure student proficiency level on the required subject/domain. The key component of CAT is "adaptive" question selection algorithm, which automatically selects best suited for based his/her current estimated proficiency, reducing length. Existing algorithms rely some manually designed pre-fixed informativeness/uncertainty metrics selections, labor-intensive not sufficient capturing complex relations between students questions. In this paper, we propose a fully adaptive framework named Neural (NCAT), formally redefines as reinforcement learning problem directly learns algorithm from real-world data. Specifically, bilevel optimization defined simplified under CAT's application scenarios make learnable. Furthermore, address task effectively, tackle it equivalent attentive neural policy model non-linear interactions. Extensive experiments datasets demonstrate effectiveness robustness NCAT compared with several state-of-the-art methods.