计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (1): 99-105.DOI: 10.3778/j.issn.1002-8331.2012-0365

• 大数据与云计算 • 上一篇    下一篇

结合学生行为模式分析的成绩早期预警研究

张明焱,杜旭,李浩   

  1. 华中师范大学 国家数字化学习工程技术研究中心,武汉 430079
  • 出版日期:2022-01-01 发布日期:2022-01-06

Research on Early Warning for Learning Performance Combined with Students’ Behavior Patterns Analysis

ZHANG Mingyan, DU Xu, LI Hao   

  1. National Engineering Research Center for E-learning, Central China Normal University, Wuhan 430079, China
  • Online:2022-01-01 Published:2022-01-06

摘要: 早期预警是在线学习中的重要主题,通过早期预警识别有不及格风险的学生可帮助教师及时开展个性化教学干预。使用深度学习模型对学生微观行为模式进行分析以提高早期预警的效果,并提出结合LSTM-autoencoder特征处理和注意力权重计算的不及格风险学生早期预警模型(LSTM-autoencoder and attention based early warning model,LAA)。该方法通过LSTM-autoencoder对学生行为时间序列数据进行特征处理,采用注意力机制计算关键预测因子。实验结果表明,LAA比基线模型取得更高的召回率,对低交互型和非持续型学生具有更好的识别效果,且能将教学干预时间提前;此外,该方法可识别影响成绩的关键周次和行为,可用于辅助教师开展在线教学指导。

关键词: 在线学习, 学习行为模式, 早期预警, LSTM-autoencoder, 注意力机制

Abstract: Early warning is an important research topic in online learning. Identifying at-risk students through early warning can help teachers conduct personalized instructional interventions. This paper applies deep learning models for micro-level learning pattern analysis to improve early warning results, LAA(LSTM-autoencoder and attentionbased early warning model) is proposed. LSTM-autoencoder is used to process the feature of students’ behavior time series, the attention mechanism is used to calculate key predictors. The case study shows that LAA achieves higher recall rate than the baseline model, has unique strength in identifying low-interaction and non-persistent students, and advances the intervention time. In addition, LAA can identify the key weeks and behaviors for instructional intervention, which can help teachers in online learning activities.

Key words: online learning, learning pattern, early warning, LSTM-autoencoder, attention mechanism