Named entity recognition for U.S. export control based on the ERNIE-BiLSTM-Attention-CRF model

Named entity recognition (NER) is a fundamental task in natural language processing, aiming to convert semi-structured or unstructured information in text into structured knowledge, so that computer systems can better understand and process text data. Due to the unclear word boundaries, complex sentence structures, and diverse word meanings in Chinese text, conventional NER methods struggle to effectively extract key information from Chinese text. To improve the accuracy and stability of the model for the task of text recognition in US export control, this paper proposes a NER method based on the ERNIE-BiLSTM-Attention-CRF model. First, ERNIE3.0 is used to preprocess text data to generate word vectors. Then, BiLSTM is used to extract key information from the context of the text. The weights of key semantic features are enhanced through the Attention fully connected layer. Finally, CRF is used to achieve the global optimal label sequence. The experiment shows that the precision, recall, and F1 value of the ERNIE-BiLSTM-Attention-CRF model on the NER task are 90.51%, 92.53%, and 91.51%, respectively, which is 13.41% higher than the baseline model CRF's F1 value, achieving good results in Chinese text recognition.