A GAUSSIAN PROCESS REGRESSION DYNAMIC PRICING ALGORITHM CONSIDERING STRATEGIC CONSUMERS

Few dynamic pricing algorithms when facing uncertain demand consider the consumers' strategic behavior. In this paper, the retailer's price decision was described as a multi-armed bandit (MAB) problem, and a non-parametric Bayesian algorithm was proposed. The algorithm combined Gaussian process regression with Thompson sampling algorithm, added strategic consumers' purchasing decision, and helped retailers to make price decisions. Simulation results show that the proposed algorithm can effectively improve retailers' revenue and converge faster. The presence of strategic consumers can improve the performance of the demand learning algorithm and reduce the loss of retailers' revenue due to the uncertainty of demand.