Simulator of reinforcement learning model in training recommendation field

The invention provides a simulator for training a reinforcement learning model in the recommendation field, which consists of two core models, namely a user state generation model based on a GAN (Generation Area Network); another is an environmental feedback algorithm, where the environmental feedback algorithm includes a user rating prediction model based on hierarchical attention, and a user feedback calculation model. According to the method, the user state can be generated, and the action generated by the recommendation agent can be fed back. Experimental results show that under the condition of a small data set, by means of the characteristics of a GAN network structure, the model can still generate an available user state, meanwhile, the scoring result is also in an available range, in addition, the constructed feedback algorithm can effectively terminate the learning process, and feedback data obtained through calculation also meets the training requirement of reinforcement learning. 本发明提供一种训练推荐领域中强化学习