A deep reinforcement learning system-based marketing cost control method, the deep reinforcement learning system comprising an agent and an execution environment, the agent being used for determining, according to a marketing strategy, a marketing action with respect to information about the state of the execution environment. Said method comprises: determining the cost of a marketing action; and then, at least according to the determined cost, determining a reward for reinforcement learning, so as to negatively correlate the reward with the cost; and thereafter, returning the reward to the agent for the agent to optimize a marketing strategy.
展开▼