We propose to generate Lagrangian cut for two-stage stochastic integer program by batch, in contrast to the existing methods which solve each Lagrangian subproblem at every iteration. We establish two convergence properties of the proposed algorithm. Then we demonstrate that the improvement in the lower bound achieved by incorporating the Lagrangian cut adheres to the 'triangle inequality', thereby showcasing the superiority of our proposed method over existing approaches. Moreover, we suggest acquiring Lagrangian cuts for unresolved scenarios by averaging the coefficients of the acquired Lagrangian cuts, ensuring the quality of this cut with a certain probability. Computational study demonstrates that our proposed algorithm can significantly improve the lower bound of the linear relaxation of the Bender master problem more quickly with much fewer Lagrangian cuts.