Thousands of dodgy copycat apps identified in Google Play Store
The researchers used neural networks to identify visually similar app icons and partially plagiarised text descriptions of the top 10,000 most popular apps in the Play Store. The ‘multi-modal embedding’ machine learning model threw up 49,608 potential counterfeits.
The potential counterfeits were then checked for malware using the private API of online malware analysis tool VirusTotal. While 7246 were tagged by at least one anti-virus tools, the researchers used a ‘relaxed threshold’ leaving them with 2040 high-risk, fake apps.
“While Google Play’s success is marked on its flexibility and customisable features that allow almost anyone to build an app, there have been a number of problematic apps that have slipped through the cracks and have bypassed automated vetting processes,” said study co-author Dr Suranga Seneviratne from the University of Sydney.