.. _sec_sentiment_cnn:
情感分析：使用卷积神经网络
==========================
在
:numref:`chap_cnn`\ 中，我们探讨了使用二维卷积神经网络处理二维图像数据的机制，并将其应用于局部特征，如相邻像素。虽然卷积神经网络最初是为计算机视觉设计的，但它也被广泛用于自然语言处理。简单地说，只要将任何文本序列想象成一维图像即可。通过这种方式，一维卷积神经网络可以处理文本中的局部特征，例如\ :math:`n`\ 元语法。
本节将使用\ *textCNN*\ 模型来演示如何设计一个表示单个文本
:cite:`Kim.2014`\ 的卷积神经网络架构。与
:numref:`fig_nlp-map-sa-rnn`\ 中使用带有GloVe预训练的循环神经网络架构进行情感分析相比，
:numref:`fig_nlp-map-sa-cnn`\ 中唯一的区别在于架构的选择。
.. _fig_nlp-map-sa-cnn:
.. figure:: ../img/nlp-map-sa-cnn.svg
将GloVe放入卷积神经网络架构进行情感分析
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
from mxnet import gluon, init, np, npx
from mxnet.gluon import nn
from d2l import mxnet as d2l
npx.set_np()
batch_size = 64
train_iter, test_iter, vocab = d2l.load_data_imdb(batch_size)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
[07:22:57] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for CPU
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
import torch
from torch import nn
from d2l import torch as d2l
batch_size = 64
train_iter, test_iter, vocab = d2l.load_data_imdb(batch_size)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
Downloading ../data/aclImdb_v1.tar.gz from http://ai.stanford.edu/~amaas/data/sentiment/aclImdb_v1.tar.gz...
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
import warnings
from d2l import paddle as d2l
warnings.filterwarnings("ignore")
import paddle
from paddle import nn
batch_size = 64
train_iter, test_iter, vocab = d2l.load_data_imdb(batch_size)
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d(X, K):
w = K.shape[0]
Y = np.zeros((X.shape[0] - w + 1))
for i in range(Y.shape[0]):
Y[i] = (X[i: i + w] * K).sum()
return Y
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d(X, K):
w = K.shape[0]
Y = torch.zeros((X.shape[0] - w + 1))
for i in range(Y.shape[0]):
Y[i] = (X[i: i + w] * K).sum()
return Y
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d(X, K):
w = K.shape[0]
Y = paddle.zeros([X.shape[0] - w + 1], dtype=X.dtype)
for i in range(Y.shape[0]):
Y[i] = (X[i: i + w] * K).sum()
return Y
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
X, K = np.array([0, 1, 2, 3, 4, 5, 6]), np.array([1, 2])
corr1d(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
array([ 2., 5., 8., 11., 14., 17.])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
X, K = torch.tensor([0, 1, 2, 3, 4, 5, 6]), torch.tensor([1, 2])
corr1d(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
tensor([ 2., 5., 8., 11., 14., 17.])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
X, K = paddle.to_tensor([0, 1, 2, 3, 4, 5, 6]), paddle.to_tensor([1, 2])
corr1d(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
W0818 09:28:13.297365 95147 gpu_resources.cc:61] Please NOTE: device: 0, GPU Compute Capability: 7.0, Driver API Version: 11.8, Runtime API Version: 11.8
W0818 09:28:13.329474 95147 gpu_resources.cc:91] device: 0, cuDNN Version: 8.7.
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
Tensor(shape=[6], dtype=int64, place=Place(gpu:0), stop_gradient=True,
[2 , 5 , 8 , 11, 14, 17])
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d_multi_in(X, K):
# 首先，遍历'X'和'K'的第0维（通道维）。然后，把它们加在一起
return sum(corr1d(x, k) for x, k in zip(X, K))
X = np.array([[0, 1, 2, 3, 4, 5, 6],
[1, 2, 3, 4, 5, 6, 7],
[2, 3, 4, 5, 6, 7, 8]])
K = np.array([[1, 2], [3, 4], [-1, -3]])
corr1d_multi_in(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
array([ 2., 8., 14., 20., 26., 32.])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d_multi_in(X, K):
# 首先，遍历'X'和'K'的第0维（通道维）。然后，把它们加在一起
return sum(corr1d(x, k) for x, k in zip(X, K))
X = torch.tensor([[0, 1, 2, 3, 4, 5, 6],
[1, 2, 3, 4, 5, 6, 7],
[2, 3, 4, 5, 6, 7, 8]])
K = torch.tensor([[1, 2], [3, 4], [-1, -3]])
corr1d_multi_in(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
tensor([ 2., 8., 14., 20., 26., 32.])
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
def corr1d_multi_in(X, K):
# 首先，遍历'X'和'K'的第0维（通道维）。然后，把它们加在一起
return sum(corr1d(x, k) for x, k in zip(X, K))
X = paddle.to_tensor([[0, 1, 2, 3, 4, 5, 6],
[1, 2, 3, 4, 5, 6, 7],
[2, 3, 4, 5, 6, 7, 8]])
K = paddle.to_tensor([[1, 2], [3, 4], [-1, -3]])
corr1d_multi_in(X, K)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
Tensor(shape=[6], dtype=int64, place=Place(gpu:0), stop_gradient=True,
[2 , 8 , 14, 20, 26, 32])
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
class TextCNN(nn.Block):
def __init__(self, vocab_size, embed_size, kernel_sizes, num_channels,
**kwargs):
super(TextCNN, self).__init__(**kwargs)
self.embedding = nn.Embedding(vocab_size, embed_size)
# 这个嵌入层不需要训练
self.constant_embedding = nn.Embedding(vocab_size, embed_size)
self.dropout = nn.Dropout(0.5)
self.decoder = nn.Dense(2)
# 最大时间汇聚层没有参数，因此可以共享此实例
self.pool = nn.GlobalMaxPool1D()
# 创建多个一维卷积层
self.convs = nn.Sequential()
for c, k in zip(num_channels, kernel_sizes):
self.convs.add(nn.Conv1D(c, k, activation='relu'))
def forward(self, inputs):
# 沿着向量维度将两个嵌入层连结起来，
# 每个嵌入层的输出形状都是（批量大小，词元数量，词元向量维度）连结起来
embeddings = np.concatenate((
self.embedding(inputs), self.constant_embedding(inputs)), axis=2)
# 根据一维卷积层的输入格式，重新排列张量，以便通道作为第2维
embeddings = embeddings.transpose(0, 2, 1)
# 每个一维卷积层在最大时间汇聚层合并后，获得的张量形状是（批量大小，通道数，1）
# 删除最后一个维度并沿通道维度连结
encoding = np.concatenate([
np.squeeze(self.pool(conv(embeddings)), axis=-1)
for conv in self.convs], axis=1)
outputs = self.decoder(self.dropout(encoding))
return outputs
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
class TextCNN(nn.Module):
def __init__(self, vocab_size, embed_size, kernel_sizes, num_channels,
**kwargs):
super(TextCNN, self).__init__(**kwargs)
self.embedding = nn.Embedding(vocab_size, embed_size)
# 这个嵌入层不需要训练
self.constant_embedding = nn.Embedding(vocab_size, embed_size)
self.dropout = nn.Dropout(0.5)
self.decoder = nn.Linear(sum(num_channels), 2)
# 最大时间汇聚层没有参数，因此可以共享此实例
self.pool = nn.AdaptiveAvgPool1d(1)
self.relu = nn.ReLU()
# 创建多个一维卷积层
self.convs = nn.ModuleList()
for c, k in zip(num_channels, kernel_sizes):
self.convs.append(nn.Conv1d(2 * embed_size, c, k))
def forward(self, inputs):
# 沿着向量维度将两个嵌入层连结起来，
# 每个嵌入层的输出形状都是（批量大小，词元数量，词元向量维度）连结起来
embeddings = torch.cat((
self.embedding(inputs), self.constant_embedding(inputs)), dim=2)
# 根据一维卷积层的输入格式，重新排列张量，以便通道作为第2维
embeddings = embeddings.permute(0, 2, 1)
# 每个一维卷积层在最大时间汇聚层合并后，获得的张量形状是（批量大小，通道数，1）
# 删除最后一个维度并沿通道维度连结
encoding = torch.cat([
torch.squeeze(self.relu(self.pool(conv(embeddings))), dim=-1)
for conv in self.convs], dim=1)
outputs = self.decoder(self.dropout(encoding))
return outputs
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
class TextCNN(nn.Layer):
def __init__(self, vocab_size, embed_size, kernel_sizes, num_channels,
**kwargs):
super(TextCNN, self).__init__(**kwargs)
self.embedding = nn.Embedding(vocab_size, embed_size)
# 这个嵌入层不需要训练
self.constant_embedding = nn.Embedding(vocab_size, embed_size)
self.dropout = nn.Dropout(0.5)
self.decoder = nn.Linear(sum(num_channels), 2)
# 最大时间汇聚层没有参数，因此可以共享此实例
self.pool = nn.AdaptiveAvgPool1D(1)
self.relu = nn.ReLU()
# 创建多个一维卷积层
self.convs = nn.LayerList()
for c, k in zip(num_channels, kernel_sizes):
self.convs.append(nn.Conv1D(2 * embed_size, c, k))
def forward(self, inputs):
# 沿着向量维度将两个嵌入层连结起来，
# 每个嵌入层的输出形状都是（批量大小，词元数量，词元向量维度）连结起来
embeddings = paddle.concat((
self.embedding(inputs), self.constant_embedding(inputs)), axis=2)
# 根据一维卷积层的输入格式，重新排列张量，以便通道作为第2维
embeddings = embeddings.transpose([0, 2, 1])
# 每个一维卷积层在最大时间汇聚层合并后，获得的张量形状是（批量大小，通道数，1）
# 删除最后一个维度并沿通道维度连结
encoding = paddle.concat([
paddle.squeeze(self.relu(self.pool(conv(embeddings))), axis=-1)
for conv in self.convs], axis=1)
outputs = self.decoder(self.dropout(encoding))
return outputs
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
embed_size, kernel_sizes, nums_channels = 100, [3, 4, 5], [100, 100, 100]
devices = d2l.try_all_gpus()
net = TextCNN(len(vocab), embed_size, kernel_sizes, nums_channels)
net.initialize(init.Xavier(), ctx=devices)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
[07:23:02] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for GPU
[07:23:03] ../src/storage/storage.cc:196: Using Pooled (Naive) StorageManager for GPU
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
embed_size, kernel_sizes, nums_channels = 100, [3, 4, 5], [100, 100, 100]
devices = d2l.try_all_gpus()
net = TextCNN(len(vocab), embed_size, kernel_sizes, nums_channels)
def init_weights(m):
if type(m) in (nn.Linear, nn.Conv1d):
nn.init.xavier_uniform_(m.weight)
net.apply(init_weights);
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
embed_size, kernel_sizes, nums_channels = 100, [3, 4, 5], [100, 100, 100]
devices = d2l.try_all_gpus()
net = TextCNN(len(vocab), embed_size, kernel_sizes, nums_channels)
def init_weights(net):
init_normal = nn.initializer.XavierUniform()
for i in net.sublayers():
if type(i) in [nn.Linear, nn.Conv1D]:
init_normal(i.weight)
init_weights(net)
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
glove_embedding = d2l.TokenEmbedding('glove.6b.100d')
embeds = glove_embedding[vocab.idx_to_token]
net.embedding.weight.set_data(embeds)
net.constant_embedding.weight.set_data(embeds)
net.constant_embedding.collect_params().setattr('grad_req', 'null')
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
glove_embedding = d2l.TokenEmbedding('glove.6b.100d')
embeds = glove_embedding[vocab.idx_to_token]
net.embedding.weight.data.copy_(embeds)
net.constant_embedding.weight.data.copy_(embeds)
net.constant_embedding.weight.requires_grad = False
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
Downloading ../data/glove.6B.100d.zip from http://d2l-data.s3-accelerate.amazonaws.com/glove.6B.100d.zip...
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
glove_embedding = d2l.TokenEmbedding('glove.6b.100d')
embeds = glove_embedding[vocab.idx_to_token]
net.embedding.weight.set_value(embeds)
net.constant_embedding.weight.set_value(embeds)
net.constant_embedding.weight.stop_gradient = True
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
lr, num_epochs = 0.001, 5
trainer = gluon.Trainer(net.collect_params(), 'adam', {'learning_rate': lr})
loss = gluon.loss.SoftmaxCrossEntropyLoss()
d2l.train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs, devices)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.093, train acc 0.967, test acc 0.869
2278.6 examples/sec on [gpu(0), gpu(1)]
.. figure:: output_sentiment-analysis-cnn_900d1d_87_1.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
lr, num_epochs = 0.001, 5
trainer = torch.optim.Adam(net.parameters(), lr=lr)
loss = nn.CrossEntropyLoss(reduction="none")
d2l.train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs, devices)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.066, train acc 0.978, test acc 0.861
4609.1 examples/sec on [device(type='cuda', index=0), device(type='cuda', index=1)]
.. figure:: output_sentiment-analysis-cnn_900d1d_90_1.svg
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
lr, num_epochs = 0.001, 5
trainer = paddle.optimizer.Adam(learning_rate=lr, parameters=net.parameters())
loss = nn.CrossEntropyLoss(reduction="none")
d2l.train_ch13(net, train_iter, test_iter, loss, trainer, num_epochs, devices)
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
loss 0.067, train acc 0.978, test acc 0.873
5186.2 examples/sec on [Place(gpu:0), Place(gpu:1)]
.. figure:: output_sentiment-analysis-cnn_900d1d_93_1.svg
.. raw:: html
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so great')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'positive'
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so bad')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'negative'
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so great')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'positive'
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so bad')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'negative'
.. raw:: html
.. raw:: html
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so great')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'positive'
.. raw:: latex
\diilbookstyleinputcell
.. code:: python
d2l.predict_sentiment(net, vocab, 'this movie is so bad')
.. raw:: latex
\diilbookstyleoutputcell
.. parsed-literal::
:class: output
'negative'
.. raw:: html
.. raw:: html
.. raw:: html
`Discussions `__
.. raw:: html
.. raw:: html
`Discussions `__
.. raw:: html
.. raw:: html
`Discussions `__
.. raw:: html
.. raw:: html