Graph Structure-based Clustering Algorithm for Android Third-party Libraries

Abstract

With the rapid development of the mobile market, the development of multi-functional applications is more efficient due to the rich functions provided by third-party libraries, so it is widely integrated into Android applications. Studies of Android third-party library security, such as hole digging, permissions, separation mechanism, the clone application test and safety test have different requirements to the third party libraries test accuracy and test center of gravity, making the Android third party libraries become a research hotspot. Besides, the detection algorithm of clustering algorithm is particularly important, therefore this article mainly research the Android third-party libraries clustering algorithm. This paper starts with the API call graph of the Android third-party library and combines the graph neural network GAT to design the similarity calculation and library clustering model of the Android third-party library. Firstly, the reverse tool was used to extract the API call diagram of the third-party Android library, and then the third-party Android library instance diagram was built based on the package dependency. Key API functions were selected to normalize the third-party Android library instance diagram, and then GAT and CNN were used as the similarity calculation model of the third-party Android library instance diagram to calculate the similarity. Finally, DBSCAN clustering algorithm is used to cluster the Android third-party library instance graph. Experimental results show that the method proposed in this paper can achieve 93% clustering accuracy and effectively cluster Android third-party libraries.