【摘要】 对用户访问行为的研究往往依托于历史访问记录,而网络管理系统统计的历史访问记录中含有大量的异常访问行为信息,严重影响了对用户访问行为规律特性的研究。当前识别方法所设定的分类器置信度低、无法有效提取异常访问行为特征,致使所设定的识别阈值难以准确识别异常访问行为。对此,提出一种基于选择性协同学习的网络域名用户异常访问行为信息精准识别方法。上述方法采用时间窗函数与Bootstrap重采样构建网络域名用户访问行为状态信息簇,利用随机加权网络的有监督学习获得访问行为状态信息模型,对模型进行稀疏化处理,获得异常访问行为信息特征。利用混合扰动生成方法建立分类器对访问行为信息样本子集进行协同学习,在学习过程中利用选择性集成进行置信度计算与访问行为信息更新,在此基础上基于准确性选取构造异常访问行为识别阈值,用于实际用户异常访问行为识别。实验结果表明,所提方法有效提高了异常访问行为信息识别精度。更多还原

【Abstract】 A method for precisely identifying information of abnormal access behavior of network domain user based on selective collaborative learning is proposed. This method used time window function and Bootstrap resampling to construct the status information cluster of access behavior of network domain user. Then, we used supervised learning of the random weighted network to obtain the information model of access behavior state, and sparsely processed the model to obtain the information characteristic of abnormal access behavior. Moreover, we used the method of mixed disturbance classifier to establish a classifier and carry out cooperative learning on the subset of access behavior information samples. In the learning process, we used selective ensemble to calculate the confidence level and update the access behavior information. On this basic, we constructed the recognition threshold value for abnormal access behavior based on accuracy selection, which was used for identification of actual access behavior of users. Simulation results prove that the proposed method effectively improves the accuracy of information recognition in abnormal accessing behavior.更多还原