APP Vetting Based on the Consistency of Description and APK

Android has witnessed a substantial growth over the years, in the market share as well as in the number of malwares. In this paper, we proposed a novel approach to detect potentially malicious applications, based on the semantic relatedness between the applications' descriptions and the apk files. We gathered an application database of 7,570 valid applications for training and testing, finding that about 16.6i¾ź% of the tested applications exhibit a lack of relatedness between the apk files and descriptions, due to either inadequate embedded text in apk file, too short a description, unsuited description, or being a malicious application. In additions, there are 4i¾ź% of applications unjustly deemed as unrelated. Our study showed that the semantic based approach is applicable in terms of malware detection and in judging the soundness of descriptions.