Many internet platforms are two-sided markets that involve two types of participants. Examples include e-commerce platforms like Taobao (retailers and consumers) and ride-hailing platforms like Uber (drivers and passengers). Participants of different types in the two-sided market have relationships (i.e., supply and demand) that provide externalities and network benefits. On two-sided platforms, marketing campaigns are designed by subsidizing supply or demand. Uplift models built in this scenario usually consider the treatment assignment for only one of the two sides. However, ignoring the interaction of treatments between two sides or treating them as noises may result in incomplete models and inaccurate predictions. As far as we know, there is not much work related to modeling the combinational treatment effects in the two-sided market. In this paper, we first introduce the two-sided treatment effects estimation problem and then propose a Unified Treatment effect Estimation (UniTE) method for one-sided and two-sided marketing. We extend the Robinson Decomposition to two-sided, in which the relationship of the three involved tasks, namely the outcome, the propensity, and the treatment effect, is theoretically derived. Based on the decomposition result, a multi-task-based neural network model is proposed to integrate the three tasks and learn the inter-task-related common information, which prompts the model to estimate the treatment effects better. We also propose a unified synthetic data generation method that adapts to one/two-sided situations to verify the treatment effects estimation performance. Extensive and comprehensive experimental results show that our method outperforms the other methods.