国际教育成就评价协会（IEA）：两步法插补结合IRT和深度学习方法用于大规模调查评估（英文版）

Large-scale assessments (LSAs) rely on matrix-sampling designs that generate substantial planned missing item-response data, necessitating robust imputation methods to support valid population inference. In this study, we report on a two-step imputation method that combines item-response-theory (IRT) and deep-learning techniques to generate a complete item-response matrix that can be used for reporting such as market-basket approach. This method consist of a variation autoencoder (VAE) initia