Mobile Application Behavior Recognition Based on Dual-Domain Attention and Meta-learning

Abstract

AbstractMobile smart devices and mobile applications carry a lot of personal information and office entertainment functions. By means of analyzing the network traffic generated by mobile applications during use, it can provide valuable information in terms of network management, privacy protection, and behavior recognition. This paper designs a recognition model based on dual-domain attention mechanism and meta-learning. First, feature extraction is performed through a deep separable convolution module, and secondly, attention is extracted from the channel and space dimensions through the attention mechanism module to enhance behavior recognition samples. At the same time, the meta-learning strategy is used for multi-task learning, so that the model can have a faster and more efficient recognition effect when facing new small sample recognition tasks. Experimental results show that compared with other small sample recognition models, the model in this paper can effectively recognize mobile application behavior.KeywordsMobile application behavior recognitionDeep learningAttention mechanismMeta-learningDeep separable convolutionSmall sample classification