基于深度学习的MOOC作弊行为检测研究

万子云; 陈世伟; 秦斌; 聂伟; 徐明

引用本文：

万子云,陈世伟,秦斌,聂伟,徐明.基于深度学习的MOOC作弊行为检测研究[J].信息安全学报,2021,6(1):32-39 [点击复制]
WAN Ziyun,CHEN Shiwei,QIN Bin,NIE Wei,XU Ming.Research on MOOC Cheating Detection Based on Deep Learning[J].Journal of Cyber Security,2021,6(1):32-39 [点击复制]

本文已被：浏览 7884次下载 7749次	码上扫一扫！
基于深度学习的MOOC作弊行为检测研究
万子云¹, 陈世伟², 秦斌³, 聂伟¹, 徐明³
0 字体:加大+\|默认\|缩小-
(1.深圳大学电子与信息工程学院深圳中国 518061;2.深圳大学机电与控制工程学院深圳中国 518061;3.深圳大学信息中心深圳中国 518061)

摘要:

快速准确地检测出MOOC学习者的作弊行为，对维护MOOC平台的发展及学习者的正常学习具有重要意义。本文研究了一种深度学习混合模型用于MOOC作弊行为的检测。该模型通过融合了卷积神经网络、双向门控循环单元以及注意力机制，大大提升了单一模型的检测性能。本文选取某MOOC平台的学习行为数据进行了实验验证，实验结果显示该模型在验证集上的精确率、召回率、AUC和误报率分别达到98.51%、81.35%、91.07%和0.016%，具有良好的应用前景。另外，本文采用了数据扩增的方法以解决MOOC作弊行为检测中存在的数据不均衡问题，实验中通过该方法进行数据平衡后，该模型在相同的验证集上的AUC提升了1.78%。

关键词: 作弊行为检测深度学习卷积神经网络双向门控循环单元注意力机制

DOI：10.19363/J.cnki.cn10-1380/tn.2021.01.03

投稿时间：2020-08-12修订日期：2020-11-16

基金项目:深圳大学和深信服科技股份有限公司广东省联合培养研究生示范基地资助；深圳大学2020年研究生教育改革项目（No.860-000001050503）资助。

Research on MOOC Cheating Detection Based on Deep Learning

WAN Ziyun¹, CHEN Shiwei², QIN Bin³, NIE Wei¹, XU Ming³

(1.School of Electronics and Information Engineering, Shenzhen University, Shenzhen 518061, China;2.School of Mechatronics and Control Engineering, Shenzhen University, Shenzhen 518061, China;3.Information Institute, Shenzhen University, Shenzhen 518061, China)

Abstract:

It is of great significance to detect cheating behaviour of MOOC learners quickly and accurately. In this paper, a hybrid model of deep learning is proposed for MOOC cheating detection. By combining Convolutional Neural Networks (CNN), Bidirectional Gated Recurrent Unit (BiGRU) and Attention mechanism, the model greatly improves the detection performance of a single model. On the data sets of a mooc platform, the experimental results show that the precision rate, recall rate, AUC and false positive rate of the model can reach 98.51%, 81.35%, 91.07% and 0.016% respectively, which have good application prospects. In addition, in order to solve the problem of data imbalance in MOOC cheating detection, the paper adopts the method of data augmentation, and the AUC of this model is improved by 1.78% by this method.

Key words: cheating detection deep learning convolutional neural networks bidirectional gated recurrent unit attention mechanism