This research introduces the DeepGraph-Net model to enhance symbol detection accuracy in Pilot-Aided OFDM Multiple Access systems. The methodology comprises of three stages: generating training data, training the model, and rigorous testing against traditional methods. The DeepGraph-Net architecture integrates Graph Neural Networks (GNNs) and Deep Long Short-Term Memory (LSTM) layers in concatenation, enabling dynamic adaptability to varying channel conditions. The dataset used for training and evaluation includes synthetic data generated using 64-QAM modulation and a Rayleigh fading channel model. Implementation details reveal a significantly reduced processing time of 0.0025 s for the DeepGraph-Net model, which achieves a Symbol Error Rate (SER) close to 10−3 at 20 dB without Cyclic Prefix (CP), outperforming benchmark methods such as Least Squares (LS) and Minimum Mean Square Error (MMSE). The results demonstrate the model’s superior performance and efficiency, making it highly suitable for real-time applications in communication systems.