Abstract: Crowdsourcing usually involves multiple parties of participants with their objectives. One of the fundamental challenges for crowdsourcing is to design effective mechanisms to make all the parties obtain their benefits besides competing. Even though fruitful previous studies have been conducted on this topic, they are usually based on the two-party crowdsourcing models under which one or more requesters and a crowd of workers are involved. However, in real-world applications, the requesters usually interact with the workers through the crowdsourcing platforms, making up the three-party crowdsourcing markets, under which the mechanism design for the requester-platform interaction doesn’t receive previous study. In this work, we model the three-party crowdsourcing market as a game with incomplete information. We show that the Nash Equilibrium of this game can be found via regret minimization with proper online learning strategies. Under the single-requester setting, we show that the classical EXP3 algorithm is optimal for the requester, meanwhile, we propose a stronger strategy for the platform based on the counterfactual regret minimization technique. We also propose effective strategies for both platform and requesters in multiple requesters setting by generalizing the single-requester strategies. The performance of the proposed strategies is verified from experiments with both synthetic and real-world datasets.