Design of English Document Automatic Extraction System Based on LSTM Network

Aiming at the problems of low processing efficiency and precision in the current English document information extraction process, LSTM (Long Short-Term Memory) is introduced to improve the accuracy and efficiency of the automatic extraction system. First, this paper builds an English document automatic extraction system based on LSTM, in which the text preprocessing module includes word segmentation, part-of-speech tagging and NER to ensure the normalization of input data. Part-of-speech tagging uses the Conditional Random Field (CRF) model, which is trained based on a large amount of data to ensure accuracy, and Named Entity Recognition (NER) uses a bidirectional LSTM-CRF model. Next, a multi-layer LSTM network is used for feature extraction to enhance semantic understanding capabilities by capturing contextual dependencies in the text. In addition, an attention mechanism is integrated to optimize the extraction of key information and increase the model’s focus on important content. Finally, a fully connected layer and classifier are used to classify and organize the extracted information to generate structured output results, verifying the system’s generalization capabilities in scientific research papers, legal documents, and business reports. The experimental results show that in the generalization ability evaluation of business reports, the precision rate is mostly concentrated around 89.5%, the recall rate is about 88%, and the F1 score is evenly maintained between 88.4% and 89.0%. The automatic extraction system based on LSTM shows significant advantages in processing complex English documents and has broad application prospects.