Machine Learning Based Recommendation System for Android Apps

The application suggestion function in current Android systems is an essential tool that users may utilise to discover a similar application to replace a targeted one. The present recommendation algorithm supplied by Google and the Google Play store supposedly suggests apps comparable to a target application while taking into account each application's popularity. However, it does not take into account the security aspects of each programme or the user's choices when doing so. A large amount and different types of mobile applications (or apps) are being offered to end users via app markets. These apps normally generate network traffic, which will consumes users' mobile data plan and may even cause potential security issues. However, the amount and type of network traffic generated by a mobile app in the wild is still poorly understood due to the lack of a systematic measurement methodology. In this paper, we first measure and analyze network traffic cost of Android apps in the official Android markets. Based on the results, we find that the apps from different categories have different traffic costs. In particular, there is a remarkable difference among the apps with similar functionality in terms of network traffic cost. Then, we add metrics of traffic cost into our app recommendation algorithm, which differs from the conventional app recommendation approaches. Experimental results show that the proposed recommendation algorithm can effectively help mobile app users avoid various potential security and privacy risks brought by the unnecessary network traffic consumption.

Keywords: Unsupervised Machine Learning, Recommendation System, Content Based Filtering.