基于注意力机制的包装命名实体识别

目的 为了解决包装行业相关文本命名实体识别困难问题,提出在BiLSTM（Bidirectional Long Short-Term Memory）神经网络中加入注意力机制（Attention）和字词联合特征,构建一种基于注意力机制的BiLSTM深度学习模型（简称Attention-BiLSTM）,以识别包装命名实体。方法 首先构建包装领域词典匹配包装语料中词语的类别特征,同时将包装语料转换为字特征和词特征联合的向量特征,并且在过程中加入POS（词性）信息。然后将以上特征联合馈送到BiLSTM网络,以获取文本的全局特征,并利用注意力机制获取局部特征。最后根据文本的全局特征和局部特征使用CRF（Conditional Random Field）解码整个句子的最优标注序列。结果 通过对《中国包装网》新闻数据集的实验,获得了85.6%的F值。结论 所提方法在包装命名实体识别中优于传统方法。

Packaging Named Entity Recognition Based on Attention Mechanism

The work aims to add attention mechanism (Attention) and Joint Characteristics of Words in BiLSTM (Bidirectional Long Short-Term Memory) neural network to construct a BiLSTM deep learning model (Attention-BiLSTM) based on attention mechanism, so as to solve the problem of difficult identification of text-related entities in the packaging industry and recognize the packaging named entity. Firstly, the packaging domain dictionary was built to match with the category features of the words in the packaging corpus, and the packaging corpus was converted into the vector features of the word feature and the character feature, and then POS (part of speech) information was added in the process. The above features were then fed jointly to the BiLSTM network to obtain the global features of the text, and the attention mechanism was used to acquire the local features. Finally, the CRF (Conditional Random Field) was used to decode the optimal label sequence of the entire sentence according to the global features and local features of the text. The final F score was 85.6% on the "China Packaging Network" news dataset. The proposed method is superior to the traditional method in packaging named entity recognition.