Research Of Android Malware Detection Model Based On Graph Neural Network

As an open-source mobile operating system based on the Linux kernel,Android’s openness is an important reason for its popularity.The openness of the Android system not only brings convenience to developers and users,but also leads to the generation of a large number of malware based on the Android platform,which seriously affects the ecological environment of the Android operating system and damages the interests of users.We analyzes the related mechanism of Android system and Android malware,and proposes two Android malware detection models based on Android function call graph and graph neural network,which improves the detection effect of Android malware,and also aims at the confrontation of Android malicious application developers.Sexual aggression also works well.The main work of this paper includes firstly,the extraction process of the function call graph is introduced,and then a semi-preserving method of reducing the function call graph is proposed,which can greatly reduce the size of the function call graph on the premise of retaining important features.Then,a method for node feature representation of Android function call graph based on library path fusion of multiple information is proposed.For adversarial attacks,this paper also proposes a function call graph adversarial attack detection method based on subgraph network(SGN),which can detect whether the sample is an adversarial attack sample through the function call graph and take corresponding defenses.Then,this paper further introduces the graph convolutional neural network framework,and then proposes a denoising graph convolutional neural network,so that the disturbance of the function call graph caused by the adversarial attack will not be diffused,and then proposes a method based on the existing The knowledge function calls the graph pooling method,and finally uses a multilayer perceptron classifier to classify the global graph vector.At the same time,considering the existence of various entity information in Android applications,this paper also proposes a detection model based on heterogeneous graph neural network,extracts various entities and relationships from Android applications,forms a heterogeneous relationship network,and uses heterogeneous Graph neural network performs the extraction and representation of hidden features in heterogeneous graphs,and constructs an Android malware detection model based on heterogeneous graph neural networks.Finally,the above two methods are experimentally verified in this paper,and similar work in recent years is selected for experimental comparison.The results show that the two models proposed in this paper have achieved better results under normal circumstances and in the presence of adversarial attacks,which proves the effectiveness and robustness of the proposed method in the face of adversarial attacks.