Identifying in-app user actions from mobile web logs

Abstract

We address the problem of identifying in-app user actions from Web access logs when the content of those logs is both encrypted (through HTTPS) and also contains automated Web accesses. We find that the distribution of time gaps between HTTPS accesses can distinguish user actions from automated Web accesses generated by the apps, and we determine that it is reasonable to identify meaningful user actions within mobile Web logs by modelling this temporal feature. A real-world experiment is conducted with multiple mobile devices running some popular apps, and the results show that the proposed clustering-based method achieves good accuracy in identifying user actions, and outperforms the state-of-the-art baseline by 17.84%.