Credit Assignment and Efficient Exploration based on Influence Scope in Multi-agent Reinforcement Learning