Ferroelectric materials are widely used in functional devices, however, achieving convenient and accurate theoretical modeling of them has been a long-standing issue. Here, we propose a noval approach for the modeling of ferroelectric materials using graph convolutional neural networks (GNN). This approach utilizes GNNs to approximate the potential energy surface of ferroelectric materials, which then serves as a calculator to enable large-scale molecular dynamics simulations. Given atomic positions, the well-trained GNN model can provide accurate predictions on the potential energy and atomic forces, reaching a level of accuracy of 1 meV, which is comparable to <i>ab initio</i> calculations but with much less time expense by orders of magnitude. Benefiting from the high accuracy and fast prediction of neural networks, we further combine it with molecular dynamics simulations to investigate two representative ferroelectric materials—bulk GeTe and CsSnI<sub>3</sub>, and successfully produce their temperature-dependent structural phase transitions, which are in good agreement with experimental observations. During the GNN-based MD simulation of GeTe, we obeserve an unusual negative thermal expansion around the region of its ferroelectric phase transition, which has been reported in previous experiments. For CsSnI<sub>3</sub>, we correctly obtain the octahedron tilting patterns associated with its phase transition sequence. These results demonstrate the accuracy and reliability of GNNs in the modeling of potential energy surfaces of ferroelectric materials, providing a universal approach for their theoretical investigations.