基于自监督神经网络在企业Web应用中的事件预测与异常检测

ENTERPRISE WEB APPLICATION EVENT PREDICTION AND ANOMALY DETECTION BASED ON SELF-SUPERVISED NEURAL NETWORKS

针对企业Web应用提出一种新的事件预测方法DeepEvent,以更好地检测异常事件。DeepEvent包括三个关键特性:特定于Web应用而设计的神经网络结构以充分考虑连续事件之间的特征,克服标注数据稀缺的自监督学习技术,以及集成上下文事件和捕获事件之间依赖关系的序列嵌入技术。在从六个真实世界的企业Web应用程序中收集到的事件上来对DeepEvent进行评估。实验结果表明,DeepEvent在预测连续事件和检测Web异常方面是有效的。

This paper proposes a new event prediction method DeepEvent for enterprise Web applications to better detect abnormal events.DeepEvent included three key features:Web-specific neural network that considered the characteristics of sequential Web events,self-supervised learning technology that could overcome the scarcity of labeled data,and sequential embedding technology that integrated contextual events and captured the dependencies among Web events.We evaluated DeepEvent on Web events collected from six real-world enterprise Web applications.Experimental results show that DeepEvent is effective in predicting sequential Web events and detecting Web anomalies.