学习分析视角下的个性化预测研究  

Predictions for personalized learning from a learning analytics perspective

作　　者：张琪 王红梅 庄鲁 赖松 Qi Zhang;Hongmei Wang;Lu Zhuang;Song Lai

机构地区：[1]江苏师范大学智慧教育学院 [2]北京师范大学教育学部教育技术学院

出　　处：《中国远程教育》2019年第4期38-45,92,93,共10页Chinese Journal of Distance Education

基　　金：教育部人文社会科学青年基金"全息数据支持的学习投入建模与干预研究"(18YJC880126)

摘　　要：以学习分析为核心的"数据驱动教学"已成为智能教育时代的重要教学范式。在学习结果预测方面,大多数研究将学习者视为整体进行评估,缺少个性化的分类形式与预测模型,也鲜有较为全面的数据挖掘算法的比较研究。本研究基于中学"互联网+"混合学习场景,基于"大五人格"分类,分析学习行为指标与不同人格特质群体学习结果之间的相关性,利用多元线性回归构建相应的预测模型。通过对28类回归算法与24类分类算法的结果比较,判别具有最佳精度和鲁棒性的预测算法。研究发现,不同人格特质群体的预测变量存在差异,课后测验平均分数出现在所有群体的预测方程中并占有最高的权重;过程性评估成绩、在线学习时长是较为稳健的预测因子;无论在数值预测还是分类预警中,RandomForest算法都具有最佳效能。Data-driven instruction, chiefly informed by learning analytics, has become an important paradigm in the era of smart education. Most studies about learning outcome prediction take students as a whole, which leads to lack of personalized classification, a personalized prediction model, as well as data mining comparison. This study, drawing from the Internet+ blended learning situations and the Big Five personality traits, analyzed the correlation between learning behaviors and learning outcomes of different personality traits, and constructed a prediction model using multiple linear regression. A prediction algorithm with the most accuracy and robustness was identified through comparing the result of 28 types of regression algorithms and 24 types of classification algorithms. The result shows that there exist prediction variables among different personality traits, with the average score of the after-class test appearing in the prediction equations of all the groups, taking up the most weight;that continuous assessment and online learning time are the two most steady prediction factors;and that random forest algorithm displays the optimal efficiency both in numeric prediction and classification early warning.

关 键 词：学习分析 学习预测 数据挖掘 人格特质 个性化建模 智能学习系统 预测效能 数据驱动教学 

分 类 号：G434[文化科学—教育学]