This paper studies the Air-Ground Collaborative Traffic Load Balancing (AGC-TLB) problem. Under such a setup, user equipment sets (UEs) perform different tasks requiring different data rates, and the macro base station (MBS) collaborates with several drone base stations (DBSs) to balance the network load condition. The user association and DBSs deployment are jointly optimized. We formulate the AGC-TLB problem as a generalized α-fairness minimization problem. Different load balancing degrees can be achieved with different coefficients α. The studied problem is a mixed-integer nonlinear programming (MINLP) problem that is challenging to be solved. To address this problem, an alternating optimization algorithm is developed. Specifically, the user association is solved by penalty-based successive convex approximation (P-SCA), whereas the DBSs deployment is modeled as an exact potential game. We analyze the existence of Nash equilibrium and propose the constrained Gibbs-sampling algorithm. Simulation results demonstrate that the proposed algorithm can achieve a load-balanced network condition compared to benchmark algorithms.