Seeds selection for influence maximization based on device-to-device social knowledge by reinforcement learning

Recently, how to use Device-to-Device (D2D) social knowledge to reduce the network traffic on mobile networks has become a hot topic. We aim to leverage D2D social knowledge to select influential users (seed users or seeds) for influence maximization to minimize network traffic. Lots of work has been done for seeds selection in a single community. However, few studies are about seeds selection in multiple communities. In this paper, we build a Multi-Community Coverage Maximization (MCCM) model to maximize the D2D social coverage so that the cellular network traffic can be minimized. We transform it into a resource allocation problem and use a Reinforcement Learning (RL) approach to tackle it. Specifically, we present a novel seeds allocation algorithm based on Value Iteration method. To reduce the time delay, we design an edge-cloud computing framework for our method by moving part of the computing tasks from the remote cloud to adjacent base stations (BSs). The experiment results on a realistic D2D data set show our method improves D2D coverage by 17.65% than heuristic average allocation. The cellular network traffic is reduced by 26.35% and the time delay is reduced by 63.53%.