基于Null Importance和Stacking模型的知识追踪研究  

A study of knowledge tracing based on Null Importance and Stacking models

作　　者：梁开迪 张丽华[1] LIANG Kaidi;ZHANG Lihua(School of Science,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学理学院,北京100876

出　　处：《中国科技论文在线精品论文》2024年第2期174-182,共9页Highlights of Sciencepaper Online

摘　　要：为应对基于游戏的学习平台在知识追踪应用方面的不足,本研究利用Field Day Lab提供的教育游戏用户日志进行深入分析。采用方差法和Null Importance方法对数据集进行降维处理,并结合K折交叉验证与LightGBM算法,建立了一个高效的预测模型。此外,通过集成Logistic模型,构建起Stacking模型。研究表明,该模型在验证集上的Macro-F1值显著提升至0.699,同时也在测试集上显示出优异的泛化能力。本研究为教育游戏领域的知识追踪提供了创新方法,并为游戏开发与教育实践提供了宝贵参考,支持教育游戏的开发者为学生创造更有效的学习体验。To address the inadequacies in knowledge tracing applications on game-based learning platforms,this study conducts an in-depth analysis using user logs from educational games provided by Field Day Lab.We applied the variance method and Null Importance method for dimensionality reduction of the dataset,and combined K-fold cross-validation with the LightGBM algorithm to develop an efficient predictive model.Furthermore,we constructed a Stacking model by integrating Logistic models.The study reveals that this model significantly improved the Macro-F1 score to 0.699 on the validation set and also demonstrated excellent generalization capabilities on the test set.This research offers innovative methods for knowledge tracing in the field of educational games and provides valuable insights for game development and educational practice,which also supports developers of educational games in creating more effective learning experiences for students.

关 键 词：人工智能 知识追踪 Null Importance Stacking集成模型 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]