Prediction Of User App Usage Behavior Based On Mobile Network Mass Traffic

In the ear of mobile network, a vast amount of mobile internet traffic data allows us to gain further insights into human activities. Researches about user behavior in mobile network are helpful to real-time service,infrastructure planning, content delivery optimization and resource optimization. Meanwhile, those researches are parts of human mobility and behavior analysis, the researches results have myriad usage in solving the problems that are closely bound up with human life and development,like urban planning, demographic change, the spread of disease.User's App usage behavior largely reflects user's behavior in mobile network. The user App usage behavior is complex and diverse. On one hand, predicting the App of a person will use in future time is critical for Internet Services Providers (ISP) to provide better personalized services,For example, it can be used in ecosystem of location-based personalized services, e.g., personalized targeted advertising, destination based contextualized reminder and itinerary recommendation. On the other hand, prediction of App usage behavior can be used in pre-loading the right Apps in memory for faster execution or pop the desired App up to the mobile's home screen, which contribute to the improvement of user experience and energy-saving.Many previous researches have proved that user mobility have significant impacts on App usage behavior. The researches in this thesis contain three aspects: advertisement traffic analysis, user mobility analysis and prediction of App usage behavior. Among them, ad traffic analysis can identity ad traffic in mobile network traffic, which is irrelevant to user App usage behavior, and the results of user mobility analysis can used to predict user App usage behavior. The contributions of this paper are summarized as follows:(1) We adopt blacklist and whitelist published by Adblock Plus (a popular advertising interceptor) as rules, and translate those rules into patterns, then identify ad traffic by pattern matching. To improve the efficiency of identifying ad traffic, we designed and implemented a fast-matching algorithm based on double index.(2) In order to reduce the influence of noise to experiment results,we use mobile network traffic without ad traffic to analysis user mobility and App usage behavior. In order to reveal the internal mechanism of App usage behavior from human mobility point of view, we focus on the impact of both crowd and individual mobility behavior on App usage.Different kinds of mobility characteristics, i.e., mobility indicators(including trip distance, radius of gyration and number of visited locations over time), time, location, travel pattern, historic App usage behavior are used to predict App usage behavior based on machine learning. The prediction accuracy achieves 91.8% for crowd and 90.3% for individual.