贝叶斯网在教育测量与因果推断中的应用——以青少年“王者荣耀”网游消费决策为例
The Educational Measurement and Causal Inference Based on Bayesian Network—Taking the Online Game Consumption Decision of “King of Glory” as an Example

贝叶斯网解决了回归技术无法处理的复杂环境下多模态数据处理，可以正向和逆向推断因果关系，将不确定性因果关系推断具象化，逐渐成为人工智能时代变量因果关系研究的关键工具。在教育测量上，贝叶斯网建立层级结构的概率模型，将先验概率和后验概率结合，提高了认知诊断的准确性，提升了计算机自适应测验的效率和效能。在因果推断上，贝叶斯网可以清晰地展现多模态数据下，多个自变量和多个因变量之间的关系，不仅突破了回归分析的局限，也使变量之间的依赖关系更加明确，使得因果推断和不确定性决策的科学性得到明显提高。采用贝叶斯网完成了516名青少年玩家“王者荣耀”消费决策影响因素的估计，发现大五人格、游戏体验、收藏偏好、产品外观和产品价格对游戏消费决策有不同的预测作用。目前贝叶斯网的概率繁殖算法还需改进，运用海量数据的结构学习和推理的问题解决方案仍存在不足，动态贝叶斯网和贝叶斯网的应用领域还需要不断扩展。
Bayesian network can solve multi-modal data processing in complex environments that cannot be processed by regression technology, and it can infer causality in both forward and reverse direc-tions, and visualize the causal inference of uncertain, which has gradually become a key tool for the research on the era variable causality of artificial intelligence. In terms of educational measure-ment, Bayesian network establishes a hierarchical probability model, combines prior probability with posterior probability, improves the accuracy of cognitive diagnosis, and improves the efficiency and efficiency of Computer Adaptive Test. In terms of causal inference, Bayesian network can clearly show the relationship between multiple independent variables and multiple dependent variables in multi-modal data, which not only breaks through the limitation of regression analysis, but also makes the dependence between variables more clear, which significantly improves the scientific nature of causal inference and uncertain decision-making. At present, the probabilistic propagation algorithm of Bayesian network still needs to be improved, and the solution to the problem of struc-tural learning and reasoning based on massive data still needs to be improved. The application fields of dynamic Bayesian network and Bayesian network need to be expanded continuously.