您的位置: 首页  >  文章  >  Recurrent Convolutional Neural Networks for Text Classification阅读笔记

Recurrent Convolutional Neural Networks for Text Classification阅读笔记

分类: 文章 • 2023-03-23 16:47:30

RCNN

Model

Recurrent Convolutional Neural Networks for Text Classification阅读笔记

Word Representation Learning

使用的双向RNN
cl(wi)c_l(w_i)表示词wiw_i左边上下文, cr(wi)c_r(w_i)表示右边上下文, 计算方式如下:

Recurrent Convolutional Neural Networks for Text Classification阅读笔记

e(wi−1)e(w_{i-1})代表第i-1个词的word embedding.
每个词的最终表示:

Recurrent Convolutional Neural Networks for Text Classification阅读笔记

再接一个全连接作为输出:

Recurrent Convolutional Neural Networks for Text Classification阅读笔记

Text Representation Learning

对上一层的输出做max-pooling

Recurrent Convolutional Neural Networks for Text Classification阅读笔记
Recurrent Convolutional Neural Networks for Text Classification阅读笔记
Recurrent Convolutional Neural Networks for Text Classification阅读笔记

实验

参数设置

optimizer: SGD
learning rate: 0.01
embedding: 50
hidden layer(y(3)的维度y^{(3)}的维度): 100

相关推荐