Influence Maximization in Near-Linear Time: A Martingale Approach