Intelligent mobile messaging for urban networks: Adaptive intelligent mobile messaging based on reinforcement learning
Behrooz Shahriari, Melody Moh · 2016
Mobile messaging has become a trend in our daily lives. The current schema for messaging is to route all the messages between mobile users through a centralized server. This scheme, though reliable, creates very heavy load on the server. It is possible for users to communicate through peer-to-peer (P2P) connection, especially over urban networks characterized by heavy user traffic and dense network connectivity. P2P connections however do not provide the best user experience, as they are sometimes unreliable due to network coverage fluctuation. We propose an intelligent messaging framework based on reinforcement learning to strike a balance between reducing server load and improving user experience. The system learns and adapts in real-time to user mobility and messaging patterns. The adaptive system dynamically chooses between routing through the server and routing via P2P connection. As it does not rely on user location information, user privacy is thus preserved. Performance evaluation through simulation of user movement and messaging patterns demonstrates that the system is able to find the best messaging policy for users, achieving a well balance between heavy server load and unreliable communication, and provides a fine user messaging experience while reducing server load. We believe that this work is significant for future smart environments and urban networking where mobile messaging will be prominent among mobile users as well as mobile smart objects.