This paper addresses the issue of poor interpretability in neural cognitive diagnosis models (CDMs) by proposing efficient Kolmogorov-Arnold networks (KANs). The authors introduce two methods to enhance the interpretability of CDMs: directly replacing multi-layer perceptrons (MLPs) with KANs and designing a novel aggregation framework using multiple KANs. The proposed model, named KAN2CND, is evaluated on four real-world datasets and shows superior performance compared to traditional CDMs and existing neural CDMs. Additionally, the learned structures of KANs in KAN2CND provide better interpretability, making it more convincing for users. The modified implementation of KANs also ensures competitive training costs. The paper's contributions include the first use of KANs to enhance CDM interpretability and the introduction of a novel aggregation framework.This paper addresses the issue of poor interpretability in neural cognitive diagnosis models (CDMs) by proposing efficient Kolmogorov-Arnold networks (KANs). The authors introduce two methods to enhance the interpretability of CDMs: directly replacing multi-layer perceptrons (MLPs) with KANs and designing a novel aggregation framework using multiple KANs. The proposed model, named KAN2CND, is evaluated on four real-world datasets and shows superior performance compared to traditional CDMs and existing neural CDMs. Additionally, the learned structures of KANs in KAN2CND provide better interpretability, making it more convincing for users. The modified implementation of KANs also ensures competitive training costs. The paper's contributions include the first use of KANs to enhance CDM interpretability and the introduction of a novel aggregation framework.