Using combined network information to predict mobile application usage

Recently, mobile applications are widely used by smartphone owners. The understanding of application usage can help us make prediction on its development tendency and meanwhile improve users' experience. To predict the future application usage, we develop a simple but novel method that considers two types of user-related networks (the users' call-log based network and the newly developed “application usage similarity network”) as well as the correlations between user characteristics and application functions. Our model is tested with data from 25,376 users selected from the largest connected component of total users in a local operator in China, and results show that the model with the combined network achieves the best result (almost 60% prediction precision with 60% recall) compared with models on a single related network (the call-log based network or the application usage similarity network) or without network information.