Everyone Deserves A Reward: Learning Customized Human Preferences