Attention-Based Bi-LSTM for Chinese Named Entity Recognition

As an integral part of deep learning, attention mechanism and bidirectional long short-term memory (Bi-LSTM) are widely used in the field of NLP (natural language processing) and their effectiveness has been well recognized. This paper adopts an attention-based Bi-LSTM approach to the question of Chinese NER (named entity recognition). With the use of word2vec, we compile vectorized dictionaries and employ Bi-LSTM models to train text vectors, with which the output eigenvectors of the attention model are multiplied. Finally, softmax is used to classify vectors in order to achieve Chinese NER. In four different configurations, our experiments describe the impact of the domain relevance of Chinese character vectors, phrase vectors, and vectorized datasets on the effectiveness of Chinese NER. The experimental results show that the standard precision (P), recall (R), and F1-score (F1) are 97.51%, 95.33%, and 96.41% respectively.