lanwenfei1@163.com
Text Classification of Chinese News Based on LSTM-Attention
投稿时间:2018-05-10  修订日期:2018-05-10
DOI:
中文关键词: 自然语言处理  深度学习  长短时记忆神经网络  注意力机制  文本分类
英文关键词: natural language processing  deep learning  Long Short-Term Memory  attention mechanism  text classification
基金项目:
作者单位E-mail
蓝雯飞 中南民族大学 计算机科学学院 lanwenfei1@163.com 
徐蔚 中南民族大学 计算机科学学院  
汪敦志 中南民族大学 计算机科学学院  
潘鹏程 中南民族大学 计算机科学学院  
摘要点击次数: 74
全文下载次数: 
中文摘要:
      经典的LSTM分类模型,一种是利用LSTM最后时刻的输出作为高一级的表示,而另一种是将所有时刻的LSTM输出求平均作为高一级的表示。这两种表示都存在一定的缺陷,第一种缺失了前面的输出信息,另一种没有体现每个时刻输出信息的不同重要程度。为了解决此问题,引入Attention机制,本文对LSTM模型进行改进,设计了LSTM-Attention模型。实验结果表明,LSTM分类模型比传统的机器学习方法分类效果更好,而引入Attention机制后的LSTM模型相比于经典的文本分类模型,分类效果也有了一定程度的提高。
英文摘要:
      The classical LSTM classification model uses the output of last moment as the higher level representation, while the other is the average representation of the output at all times as a higher level. Both of these representations have some defects, the first one missing the previous output information, the other does not reflect the different importance of the output information at each moment. In order to solve this problem, the attention mechanism has been introduced to improve the LSTM model, and the LSTM-attention model is designed. The experimental results show that the LSTM classification model is better than the traditional machine learning method, and the LSTM model has improved to some extent compared with the classical text categorization model after introducing the attention mechanism.
View Fulltext   查看/发表评论  下载PDF阅读器
关闭