基于字符与单词嵌入的航空安全命名实体识别

Named Entity Recognition Based on Character and Word Embedding in Aviation Safety

航空安全命名实体识别是构建航空安全知识图谱中基础且关键的任务,对消除航空隐患,制定有效的纠正措施和宏观政策提供了重要依据。针对航空安全领域包含大量较长的专有名词和名词缩写混合等问题,采用双向长短期记忆模型(BILSTM)、卷积神经网络(CNN)和条件随机场(CRF),构建一种使用字符与词两个粒度的模型,对航空安全事故进行命名实体识别(NER),以提取事故中的实体。采用航空事故报道为实验数据集,利用BILSTM模型自动学习字符粒度的语义特征向量,再结合词粒度的特征向量,通过CNN全局特征,最后通过CRF层对提取到的特征进行序列标注,以提取命名实体。经过实验对比验证,该模型能够有效提取命名实体,F1值相对现有方法提升了2.22%。实验结果表明,增加字符粒度的嵌入并且使用CNN获取全局特征可以有效提高航空安全领域命名实体识别效果。

The identification of aviation safety nomenclature is a basic and critical task in the construction of aviation safety knowledge map,which provides an important basis for eliminating aviation hidden dangers and formulating effective corrective measures and macro policies.Aimed at locally sensitive problems such as the mixture of a large number of long proper nouns and noun abbreviations,we use bidirectional long-term short-term memory model(BILSTM),convolutional neural network(CNN)and conditional random field(CRF)to construct a model that uses two granularities of characters and words to perform named entity recognition(NER)for aviation safety accidents to extract the entity in the accident.The aviation accident report is used as the experimental data set,and the BILSTM model is used to learn the semantic feature vector of the character granularity.Combined with the feature vector of the word granularity,the global feature is obtained through CNN,and finally the extracted features are sequenced through the CRF layer to extract the name entity.After experimental comparison and verification,the model can effectively extract named entities and the F1 value is increased by 2.22%compared with the existing methods.Experimental results show that increasing the embedding of character granularity and using CNN to obtain global features can effectively improve the effect of named entity recognition in the aviation safety field.