A novel multiuser digital semantic communication (D-SemCom) framework is proposed to enhance the adaptability and throughput of multiple-input multiple-output (MIMO)orthogonal frequency division multiplexing (OFDM) systems by integrating the non-orthogonal superposition pilot (NOSIP) scheme. However, this introduces challenges, such as inter-user interference, which limit the performance of traditional signal processing methods. To address these challenges, we propose SANet, which realizes the nonlinear mapping of received signals to log-likelihood ratios. SANet leverages a common subspace and a multi-head self-attention mechanism to effectively suppress interference and optimize signal decoding performance. Additionally, SANet addresses the coupling between the semantic encoder-decoder and the channel environment, reducing the dependency on channel variations and enabling the system to adapt more flexibly to different channel and data scenarios. We also introduce an adaptive data augmentation strategy that dynamically adjusts the data representation based on channel conditions, further enhancing the reliability and performance of D-SemCom. Numerical results show that: 1) SANet improves the throughput of D-SemCom by 7.01% compared to traditional approaches; 2) the throughput of D-SemCom is improved by 37.15% with the NOSIP scheme compared to the orthogonal pilot scheme; and 3) the effectiveness of the adaptive data augmentation strategy in enhancing D-SemCom performance is validated.