On the choice of regularization parameter when solving ill-posed problems in reflexive spaces☆