BiLSTM-based with Word-weight Attention for Chinese Named Entity Recognition

Abstract

Natural language processing is a hot research area in recent years. Named entity recognition is a fundamental task in natural language processing. However, Chinese named entity recognition usually suffers from weak relationship between related words and sentences resulting in recognition errors. In order to clarify the weight of different words in sentences and strengthen the dependence between character and words, we propose a named entity recognition model LSTM-WWAT based on bidirectional long-term memory network (BiLSTM) and word-weight attention(WWAT). Firstly, we add word semantic information into the character vector of the embedding layer by matching the dictionary. Secondly, we use the BiLSTM to extract the context dependent features of characters and related words. Then, the model import the hidden vector into WWAT and depend on sentences features to strengthen the word weight, so that the output will be closer to the entity annotation we want. Finally, Random Conditional Field (CRF) is used to decode the optimal coding sequence as the result of named entity recognition. Experimental results show that, compared with baseline models, our model achieves significant improvements.