The Expectation-Maximization (EM) algorithm is widely used for parameter estimation in item response theory (IRT) modeling. Despite its utility, when applied to datasets with large numbers of individuals and items, the standard EM algorithm can be slow to converge, with computationally expensive E-steps. This study proposes a modified EM algorithm with a two-stage structure that capitalizes on data subsets to accelerate estimation for unidimensional two-parameter logistic (2PL) IRT models. Two simulation studies evaluated the reduction in convergence time, bias and root mean squared error (RMSE) of parameter recovery, and bias and RMSE of standard error (SE) estimation under the proposed algorithm. The proposed algorithm was also compared to standard EM across varying subset sizes and item set lengths. Results showed that (a) parameter recovery differed between types of IRT parameters and between moderate and extreme parameter values; (b) recovery improved with larger subset sizes; (c) reduction in time was more pronounced with a larger item set; and (d) a small upward bias and RMSE in SE estimates were observed compared to standard EM, though both metrics remained modest in all conditions. Overall, the proposed algorithm improved computational efficiency while maintaining comparable estimation accuracy and precision under larger item sets and moderate-to-large subset sizes. The study concludes with future applications in large-scale operational contexts with large volumes of examinees and moderate to large item pools, and potential extensions to multidimensional IRT settings.