基于反向选择的网络异常学习行为识别方法  

An online anomaly learning behavior detection method based on negative selection algorithm

作　　者：杨鹤[1,2,3] 彭璐 刘清堂 杨莉[1] 雷建军[1] YANG He;PENG Lu;LIU Qingtang;YANG Li;LEI Jianjun(School of Computer&Hubei Education Cloud Service Engineering Technology Research Center,Hubei University of Education,Wuhan 430205,China;Wuhan Huada National E-Learning Technologies Co.,Ltd,Central China Normal University,Wuhan 430079,China;National Engineering Research Center for E-Learning,Central China Normal University,Wuhan 430079,China;School of Computer,Wuhan University,Wuhan 430064 China)

机构地区：[1]湖北第二师范学院计算机学院&湖北省基础教育信息技术服务协同创新中心,武汉430205 [2]华中师范大学武汉华大数字化学习工程技术有限公司,武汉430079 [3]华中师范大学国家数字化学习工程技术研究中心,武汉430079 [4]武汉大学计算机学院,武汉430064

出　　处：《中南民族大学学报（自然科学版）》2023年第5期664-671,共8页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：中国博士后基金资助项目(2018M640738);湖北省高等学校省级教学研究资助项目(2020674);湖北省教育厅重点科学研究资助项目(D20193002)。

摘　　要：网络学习中,异常学习行为不易及时被察觉和纠正,可能会导致严重的学习问题.网络异常学习行为具有多样性和不确定性,难以通过规则直接界定.借鉴生物免疫系统识别病原体的原理设计的反向选择算法,能自适应识别未知异常,并具有实时性、动态性、多样性、鲁棒性等特征.借助主成分分析法从网络学习行为日志数据中抽取行为特征,构成多维空间的学习行为向量,通过优化训练集改进了反向选择算法并设计了基于该算法的网络异常学习行为识别方法.在真实数据集上的实验结果表明:该方法的识别率优于朴素高斯贝叶斯、决策树、支持向量机等常用算法,能够及时对异常学习行为进行早期预警,为干预和改进学习效果提供客观依据.该方法不需要人工干预,能识别未知的异常行为,具有多样性和较高的自适应性.In online learning,abnormal learning behavior is difficult to be detected and corrected in time,which may lead to serious learning problems.Online abnormal learning behavior is diverse and uncertain,and difficult to be distinguished by rules.The Negative Selection Algorithm inspired from biological immune system to recognize pathogens could adaptively detect unknown abnormalities and has the characteristics of real-time,dynamism,diversity and robustness.With the help of Principal Component Analysis Method,behavioral features are extracted from from online learning behavior log data to constitute a multi-dimensional space of learning behavior vectors.By optimizing the training set,the Negative Selection Algorithm is improved and online abnormal learning behavior detecting method based on this algorithm is designed.Experimental results on real datasets show that the accuracy rate of the method is better than that of common algorithms such as Bayes,Decision Trees and SVM.It can give early warning to abnormal learning behavior in time and can provide an objective basis for intervention and improvement of learning achievement.In addition,this method does not need human intervention and can identify unknown abnormal behavior with diversity and high adaptability.

关 键 词：反向选择算法 人工免疫系统 学习行为识别 异常学习行为 

分 类 号：TP3.5[自动化与计算机技术—计算机科学与技术] G40.57[文化科学—教育学原理]