基于Attention-BiLSTM的中文命名实体识别

提出一种基于Attention-BiLSTM(attention-bidirectional long short-term memory)深度神经网络的命名实体识别方法。应用BiLSTM神经网络自动学习文本的隐含特征,可以解决传统识别方法存在长距离依赖等问题;引入注意力机制(attention mechanism)对文本全局特征做重要度计算,获取文本局部特征,解决了传统深度学习方法不能充分提取特征的问题;在预训练过程中加入维基百科知识,进一步提升了命名实体识别系统的性能。实验表明,所提方法在SIGHAN 2006 Bakeoff-3评测数据集上获得了优良的识别性能。

Entity Recognition of Chinese Names Based on Attention-BiLSTM

This paper proposes a named entity recognition method based on Attention-BiLSTM (attention-bidirectional long short-term memory) deep neural network. Using BiLSTM neural network to automatically learn the implicit features of text can solve the problem of long-distance dependence of traditional recognition methods. Attention mechanism is used to calculate the importance of text global features, obtain local features of text, and solve the traditional deep learning method can not fully extract the feature problem; adding Wikipedia knowledge in the pre-training process further improves the performance of the named entity recognition system. Experiments show that the proposed method achieves excellent recognition performance on the SIGHAN 2006 Bakeoff-3 evaluation data set.