With the proliferation of electric vehicles (EVs), coordinating Vehicle-to-Grid (V2G) technology in distributed networks can enhance grid flexibility. However, the uncertainties of EVs pose big challenges to the implementation of V2G. In particular, the uncertain charging demand of EVs is impacted by real-time prices, i.e., decision-dependent uncertainty (DDU). To this end, this paper proposes a bilevel game framework considering DDUs for V2G regulation in a distribution network. A chance-constrained learning (CCL) method is proposed to capture the probability distributions of DDUs under different pricing decisions. This relaxes the long-existing assumption of the pre-known relationship between DDUs and decisions. Instead, it directly constructs these distributions from data, which gives more accurate modeling results. Due to the incorporation of the neural network model into the optimization constraints by CCL, the original problem is transformed into a mixed-integer quadratic constrained problem (MIQCP) with bilinear terms. Then, an improved segmentation-based approximation method is introduced to reformulate the original problem as an MIQCP with linear terms, which makes the problem solvable. Finally, the IEEE 123-bus system is utilized to validate the performance of the proposed method in terms of both computational speed and solution accuracy.