The concern of this paper is the numerical solution of the moder-ately ill-posed first kind integral equations in reproducing kernel Hilbert spaces (RKHS). Different with previous works in [9, 29], the amount of observation data is large, and corrupted by randomly distributed noise with large variance. In this paper, a purely data driven semi-discrete Tikhonov regularization method is proposed, effectively reconstructing the sought solution. The choice rules of parameters are provided and the rigorous upper bound estimation of confidence interval of the error in L2norm is established. Some numerical examples are provided to illustrate the appropriateness of the parameter choice and show the computational performances of the method.