Mining the most influential k-location set from massive trajectories