中国电化教育

随着教育数字化与智能化的发展,认知诊断模型在教学评价与教学干预中的作用日益凸显。现有研究多以提升诊断精度为主要目标,通过引入复杂模型结构增强表达能力,一定程度上忽视了教学系统在实际运行过程中所面临的计算资源约束与部署运行成本问题,限制了模型在真实教学场景中的应用。针对上述问题,该文从教学系统实际需求出发,将认知诊断模型构建问题表述为一个面向教学场景的资源约束建模优化问题,提出了一种资源约束条件下的认知诊断模型自动构建方法。该方法在保证基本诊断性能的前提下,将计算资源消耗显式纳入模型构建目标,通过多目标联合评价机制,在诊断性能与计算复杂度之间实现合理平衡,并引入群体搜索思想对模型结构进行自动生成与筛选,避免对固定模型形式的人工依赖。在ASSISTments与SLP等真实教学数据集上的实验结果表明,该文方法在显著降低模型推理计算量的同时,仍能保持具有竞争力的诊断性能,并在不同场景下表现出较好的稳定性与适应性。该研究为认知诊断模型在资源受限教学系统中的规模化应用提供了一种具有实践价值的建模思路。

With the advancement of educational digitalization and intelligent learning systems, cognitive diagnosis models(CDMs) have played an increasingly important role in educational assessment and instructional intervention. However, most existing studies primarily focus on improving diagnostic accuracy by introducing complex model structures, while largely overlooking the computational resource constraints and deployment costs faced by real-world instructional systems. This limitation hinders the large-scale application of cognitive diagnosis techniques in practical educational scenarios. To address this issue, this paper reformulates the construction of cognitive diagnosis models as a resource-constrained optimization problem oriented toward instructional system requirements. An adaptive automatic construction method for cognitive diagnosis models under computational resource constraints is proposed. While ensuring fundamental diagnostic effectiveness, computational cost is explicitly incorporated into the model construction objectives through a multi-objective evaluation mechanism, enabling a systematic trade-off between diagnostic performance and computational complexity. Furthermore, a population-based search strategy is employed to automatically generate and select model structures, reducing reliance on manually predefined model forms. Experimental results on real-world educational datasets, including ASSISTments and SLP, demonstrate that the proposed method can significantly reduce inference computation while maintaining competitive diagnostic performance, and exhibits robust stability and adaptability across different application scenarios. This study provides a practical and system-oriented modeling paradigm for the scalable deployment of cognitive diagnosis models in resource-constrained educational systems.