Title:AndroVault: Constructing Knowledge Graph from Millions of Android Apps for Automated Analysis

Abstract:Data driven research on Android has gained a great momentum these years. The abundance of data facilitates knowledge learning, however, also increases the difficulty of data preprocessing. Therefore, it is non-trivial to prepare a demanding and accurate set of data for research. In this work, we put forward AndroVault, a framework for the Android research composing of data collection, knowledge representation and knowledge extraction. It has started with a long-running web crawler for data collection (both apps and description) since 2013, which guarantees the timeliness of data; With static analysis and dynamic analysis of the collected data, we compute a variety of attributes to characterize Android apps. After that, we employ a knowledge graph to connect all these apps by computing their correlation in terms of attributes; Last, we leverage multiple technologies such as logical inference, machine learning, and correlation analysis to extract facts (more accurate and demanding, either high level or not, data) that are beneficial for a specific research problem. With the produced data of high quality, we have successfully conducted many research works including malware detection, code generation, and Android testing. We would like to release our data to the research community in an authenticated manner, and encourage them to conduct productive research.