摘要
针对神经网络文本分类模型随着层数的加深,在训练过程中发生梯度爆炸或消失以及学习到的词在文本中的语义信息不够全面的问题,该文提出了一种面向中文新闻文本分类的融合网络模型。该模型首先采用密集连接的双向门控循环神经网络学习文本的深层语义表示,然后将前一层学到的文本表示通过最大池化层降低特征词向量维度,同时保留其主要特征,并采用自注意力机制获取文本中更关键的特征信息,最后将所学习到的文本表示拼接后通过分类器对文本进行分类。实验结果表明:所提出的融合模型在中文新闻长文本分类数据集NLPCC2014上进行实验,其精度、召回率、F1-score指标均优于最新模型AC-BiLSTM。
To avoid the issue of gradient disappearance or gradient explosion associated with the deeper layers and better capture the word semantic information, this paper proposed a fusion network for Chinese news text classification. Firstly, this paper applies the densely connected bi-GRU to learn the deep semantic representation. Secondly, it applies max-pooling layer to reduce the key vector dimension. Thirdly, it adopted the self-attention mechanism to capture more important features. Finally, the learning representations are concatenated as the input of the classifier. The experimental results on NLPCC2014 dataset show that the proposed fusion model is better than the latest model AC-BiLSTM.
作者
胡玉兰
赵青杉
陈莉
牛永洁
HU Yulan;ZHAO Qingshan;CHEN Li;NIU Yongjie(Department of Computer,Xinzhou Teachers University,Xinzhou,Shanxi 034000,China;School of Information Science&Technology,Northwest University,Xi'an,Shaanxi 710127,China;School of Mathematics&Computer Science,Yan'an University,Yan'an,Shaanxi 716000,China)
出处
《中文信息学报》
CSCD
北大核心
2021年第3期107-114,共8页
Journal of Chinese Information Processing
基金
国家重点研发项目(2017YFB402103-1)。
关键词
文本分类
密集连接
双向门控循环神经网络
最大池化
自注意力机制
text classification
dense connection
bi-direction gated recurrent unit
max pooling
self-attention mechanism