1: \begin{abstract}

2: Complex social systems are composed of interconnected individuals

3: whose interactions result in group behaviors. Optimal control of a

4: real-world complex system has many applications, including road traffic management, epidemic prevention, and information dissemination.

5: However, such real-world complex system control is difficult to achieve

6: because of high-dimensional and non-linear system dynamics, and the

7: exploding state and action spaces for the decision maker. Prior methods

8: can be divided into two categories: simulation-based and analytical approaches.

9: Existing simulation approaches have high-variance in Monte Carlo integration, and the

10: analytical approaches suffer from modeling inaccuracy. We adopted

11: simulation modeling in specifying the complex dynamics of a complex

12: system, and developed analytical solutions for searching optimal strategies

13: in a complex network with high-dimensional state-action space. To

14: capture the complex system dynamics, we formulate the complex social network

15: decision making problem as a discrete event decision process. To address

16: the curse of dimensionality and search in high-dimensional state

17: action spaces in complex systems, we reduce control of a complex system

18: to variational inference and parameter learning, introduce Bethe entropy

19: approximation, and develop an expectation propagation algorithm. Our

20: proposed algorithm leads to higher system expected rewards, faster

21: convergence, and lower variance of value function in a real-world

22: transportation scenario than state-of-the-art analytical and sampling

23: approaches.

24: \end{abstract}

25: