Attention-Based Bidirectional Long Short-Term Memory Networks for Chinese Named Entity Recognition

Abstract

Named entity recognition is an important task in natural language processing and has been carefully studied in recent decades. In this paper, we investigate the problem of Chinese named entity recognition. Using attention mechanisms based on BiLSTM-CRF model, a model is proposed in this paper, which makes better use of word-based and character-based information. All the potential words that match the input characters and sentences with the dictionary are encoded, and one attention layer to control the dynamic acquisition of multiple potential characters in different paths from sequence information. A series of input characters and all potential words matched with dictionaries in sentences are encoded to measure the correlation scores between candidate characters and potential words. Another attention layer is to produce a weight vector and merge word-level features from each time step into a sentence-level feature vector by multiplying the weight vector. Then, CRF model is introduced to get the final tagging to obtain the desired result. The experimental data shows that the F1-score of our model has increased from 73.88% to 75.10% on the OntoNote 4 dataset, and from 93.18% to 94.17% on the MSRA dataset. The results show that our method has a better performance than the previous model.