面向服务等级的网络流多任务分类方法

Multitask learning for class-of-service oriented network traffic classification

在网络流分类实践中,网络运营商通常只需要知道网络流所需的服务类别(class of service,CoS),就可对网络流优先级和资源分配做出决定。为了满足用户对体验质量的需求,提出了面向服务等级的网络流多任务分类方法。该方法是直接进行面向CoS的流分类,而不需要推断应用类型。同时提出多任务框架,利用领域知识定义宏特征组及应用合作博弈中的Shapley Value模型来合理分析特征,并用决策树分箱来解决CoS阈值划分问题。采用真实网络数据集进行实验,通过在少量标记数据的情况下,优化网络参数和调整各网络模型时间损耗和分类准确性的稳定相关系数。结果表明,该方法分类准确度(提高了12.66%)和时间消耗(减少了39.23%)性能优于现有文献方法,同时分析了多分类实验结果并给出有关建议。

In the practice of network flow classification,network operators usually only need to know the class of service(CoS)required by the network flow,and can make decisions on the priority of network flow and resource allocation.Therefore,this paper designs a class-of-service network flow classification method to meet the user’s demand for quality of experience,which is based on multi-task learning.This method performs CoS classification directly without inferring the application type.At the same time,the paper proposes a multi-task framework that uses the following two ways to analyze features in a rational way:(1)domain knowledge to define macro-feature groups and application of the Shapley Value model in cooperation and games;(2)a decision tree splitting box to solve the CoS threshold division problem.Experiments are carried out on real network datasets,by optimizing network parameters and adjusting the stable correlation coefficient of time loss and classification accuracy of each network model in the case of a small amount of labeled data.The results show that the accuracy(improved by 12.66%)and time consumption(decreased by 39.23%)of the proposed method are better than those of the existing methods.The results of multi-classification experiments are also analyzed and relevant suggestions are given.