基于TextCNN-Attention-BiLSTM融合模型的煤矿隐患文本分类研究
Research on Classification of Coal Mine Hidden Danger Texts Based on TextCNN-Attention-BiLSTM Fusion Model
罗海平 1曾向阳 1陈勇1
作者信息
- 1. 武汉科技大学 资源与环境工程学院,湖北 武汉 430081
- 折叠
摘要
为实现大量煤矿隐患文本的迅速、精确分类,及时了解安全概况并加以管理.首先,选取安全文库网中多个煤矿隐患数据库为实验数据源,对煤矿隐患文本进行预处理,包括去除噪声词、分词和词向量表示等;其次,利用TextCNN对文本进行卷积操作,提取不同尺寸的特征表示,再利用BiLSTM模型对得到的特征向量进行时序建模,并结合注意力机制(Attention),从而更好地关注文本中关键信息,捕捉文本全局语义信息;最后,利用全连接层的多标签分类器预测文本隐患类别.实验结果表明:TextCNN-Attention-BiLSTM融合模型在准确率、精确率、召回率和F1 值上均达到92%以上,为煤矿隐患文本分类提供了一种更加准确和有效的解决方案,对煤矿安全管理优化具有重要意义.
Abstract
In order to achieve quick and accurate classification of a large number of coal mine hazard texts,and timely un-derstand the safety situation for effective management,first,multiple coal mine hidden danger databases from the website of safety library were selected as experimental data sources,the coal mine hidden danger texts were preprocessed,including noise words removal,word segmentation,and word vector representation,etc.Then,the TextCNN(text convolutional neural network)was used to perform convolution operations on the texts,extracting feature representations of different sizes.The BiLSTM(bi-direc-tional long short-term memory)model was utilized to sequentially model the obtained feature vectors.Combined with the atten-tion mechanism(Attention),the model can better focus on key information in the texts and capture the global semantic informa-tion of the texts.Finally,a multi-label classifier in the fully connected layer was used to predict the categories of hidden dan-gers in the texts.Experimental results showed that the fused TextCNN-Attention-BiLSTM model achieves over 92%accuracy,precision,recall,and F1 value,providing a more accurate and effective solution for coal mine hidden danger text classification.It is of great significance for optimizing coal mine safety management.
关键词
煤矿安全/TextCNN/注意力机制/BiLSTM/文本分类Key words
coal mine safety/TextCNN/attention mechanism/BiLSTM/text classification引用本文复制引用
出版年
2024
NETL
NSTL
维普
万方数据