Game and Reference: Efficient Policy Making for Epidemic Prevention and Control

Abstract

Epidemic policy-making, as a special data-mining task, is proposed to predict the proper intensities of certain epidemic prevention and control policies based on the spatial-temporal data related to regional epidemics.Previous studies are currently constrained by two issues: First, existing methods are all strongly supervised by policy effect evaluation, since only a small proportion of factors in real-world policy-making are modeled, policies made by the existing models are then easily become extreme or unreasonable. Second, the subjectivity and the cognitive limitation of humans make historical policies not always optimal for the training of decision models. To this end, we present a novel Policy Combination Synthesis (PCS) model for epidemic policy-making. In particular, to prevent extreme decisions, we introduce adversarial learning between the model-made policies and the real policies to force the output policies to be more human-like. On the other hand, to minimize the impact of sub-optimal historical policies, we employ contrastive learning to let the model draw on experience from the best historical policies under similar scenarios. Both adversarial learning and contrastive learning are adaptive to the comprehensive effects of real policies, therefore ensuring that the model always learns useful information.Extensive experiments on real-world data show that policies made by the proposed model outperform the baseline models on both the epidemic containment effect and the economic impact, thereby proving the effectiveness of our work.