Variational Bayesian inference for the multiple-choice DINA model
Variational Bayesian inference for the multiple-choice DINA model
复制标题
多项选择 DINA 模型的变分贝叶斯推理
DOI：
10.1007/s41237-020-00104-w
复制
发表时间：
2020
期刊：
影响因子：
--
通讯作者：
K.
中科院分区：
文献类型：
--
作者：
Yamaguchi;K.
在这项研究中，已提出了一个完全贝叶斯公式的多项选择的项目版本的确定性输入噪声“与”门（MC-DINA）模型，这是一个认知诊断模型，用于从多项选择的响应数据中提取信息。此外，一个变分推理算法包含一个经验贝叶斯估计过程，以解决沉重的计算负担问题的贝叶斯统计过程。所提出的方法是一样快的期望最大化算法，因为它不需要生成随机数（不像马尔可夫链蒙特卡罗技术）。该算法通过极大化边缘似然函数的对数下界，自动从分析数据中提取最优超参数。仿真结果表明，该方法可以成功地恢复真实的项目和学生的参数。
In this study, a fully Bayesian formulation has been proposed for the multiple-choice item version of the deterministic input noisy “AND” gate (MC-DINA) model, which represents a cognitive diagnostic model for extracting information from multiple-choice response data. In addition, a variational inference algorithm containing an empirical Bayesian estimation procedure was developed to solve heavy computational burden problems in Bayesian statistics procedure. The proposed method is as fast as the expectation–maximization algorithm because it does not require generation of random numbers (unlike the Markov chain Monte Carlo technique). Moreover, this algorithm can automatically extract optimal hyperparameters from analyzed data by maximizing the lower bound of the logarithm of the marginal likelihood function. The results of simulations showed that the proposed technique could successfully recover the true item and student parameters.