Identifying influential agents is an important issue in controlling the dynamical processes in complex networks, while centrality measurements are good ways to rank node\'s influence. In this paper, we combine agents\' influence with their strategy-updating timescales to explore the effect of diverse influences on the emergence of cooperation in the evolutionary prisoner\'s dilemma game. Through Monte Carlo simulation in five real-world social networks, we find that collective influence outperforms the basic centrality measurements such as degree and coreness in ranking agent\'s influence, identifying the influential spreaders and promoting the evolution of cooperation. Moreover, collective influence with depth length plays an nontrivial role on the evolution of cooperation in networks.