Design of English Document Automatic Summarization and Translation System using Long Short-Term Memory Network

This paper proposes an end-to-end system based on long short-term memory (LSTM) to address the limitations of traditional methods in handling long-distance dependencies and complex semantic relationships in automatic summarization and translation. The system preprocesses documents by word segmentation, stop word removal, and stem extraction, converting words into low-dimensional vectors via word embedding. A bidirectional LSTM (Bi-LSTM) encoder captures contextual semantics to generate fixed-length semantic vectors. The decoder integrates an attention mechanism to dynamically focus on relevant input information and uses the Beam Search algorithm to optimize the coherence and accuracy of generated summaries or translations. The model is trained on large-scale public corpora, with parameters optimized using cross-entropy loss and mini-batch gradient descent. Experimental results show that the system achieves a ROUGE score of 0.78 in summarization and a BLEU score of over 0.83 in translation, with high accuracy, fluency, and diversity. This study provides new technical support for automatic summarization and translation, advancing natural language processing.