Enhancing named entity recognition from military news with bert
Hao Qin, Yongli Wang · Journal of Physics Conference Series · 2020
Abstract Information extraction from news pages is often used in various fields. Named entity recognition is a key step in information extraction. News data is difficult to process because of it has unstructured characteristics. So the author proposes a method to perceive the relationship between words and sentences. The method is integrated with BERT(Bidirectional Encoder Representations from Transformers)-BiLSTM(Bi-directional LongShortTerm Memory)-CRF(Conditional Random Field), referred to as the Bertbc model. The model was tested on the People’s Daily dataset and text data from military news. The results show that this method improves the recognition accuracy, recall rate, f value and recognition effect.