Research on Administrative Management Optimization Strategy Based on LSTM and Transformer

With the increasing complexity of Administrative Management (AdMa) processes, improving efficiency through effective optimization strategies has become a critical research topic. This paper proposes an AdMa optimization model based on the integration of LSTM and Transformer to enhance process efficiency using deep learning techniques. Specifically, a hybrid model is constructed: the LSTM module captures temporal dependencies in sequential process steps to generate preliminary time predictions, while the Transformer module extracts global contextual features from process text descriptions. By concatenating the outputs of both models and feeding them into a fully connected layer, the final optimized process time prediction is obtained. Experiments demonstrate that the proposed model achieves the lowest Mean Absolute Error (MAE) of 0.54 h and 0.56 h on two distinct datasets, with time optimization rates reaching $16.78 \%$ and $17.13 \%$, respectively. These results validate the model’s effectiveness in real-world AdMa optimization applications.