PyTorch教程-16.1. 情緒分析和數(shù)據(jù)集

隨著在線社交媒體和評論平臺的激增，大量的意見數(shù)據(jù)被記錄下來，具有支持決策過程的巨大潛力。情感分析研究人們在其生成的文本中的情感，例如產(chǎn)品評論、博客評論和論壇討論。它在政治（例如，公眾對政策的情緒分析）、金融（例如，市場情緒分析）和市場營銷（例如，產(chǎn)品研究和品牌管理）等領(lǐng)域有著廣泛的應(yīng)用。

由于情緒可以被分類為離散的極性或尺度（例如，積極和消極），我們可以將情緒分析視為文本分類任務(wù)，它將可變長度的文本序列轉(zhuǎn)換為固定長度的文本類別。在本章中，我們將使用斯坦福的大型電影評論數(shù)據(jù)集進(jìn)行情感分析。它由一個訓(xùn)練集和一個測試集組成，其中包含從 IMDb 下載的 25000 條電影評論。在這兩個數(shù)據(jù)集中，“正面”和“負(fù)面”標(biāo)簽的數(shù)量相等，表明不同的情緒極性。

import os
import torch
from torch import nn
from d2l import torch as d2l

import os
from mxnet import np, npx
from d2l import mxnet as d2l

npx.set_np()

16.1.1。讀取數(shù)據(jù)集

首先，在路徑中下載并解壓這個 IMDb 評論數(shù)據(jù)集 ../data/aclImdb。

#@save
d2l.DATA_HUB['aclImdb'] = (d2l.DATA_URL + 'aclImdb_v1.tar.gz',
             '01ada507287d82875905620988597833ad4e0903')

data_dir = d2l.download_extract('aclImdb', 'aclImdb')

Downloading ../data/aclImdb_v1.tar.gz from http://d2l-data.s3-accelerate.amazonaws.com/aclImdb_v1.tar.gz...

#@save
d2l.DATA_HUB['aclImdb'] = (d2l.DATA_URL + 'aclImdb_v1.tar.gz',
             '01ada507287d82875905620988597833ad4e0903')

data_dir = d2l.download_extract('aclImdb', 'aclImdb')

Downloading ../data/aclImdb_v1.tar.gz from http://d2l-data.s3-accelerate.amazonaws.com/aclImdb_v1.tar.gz...

接下來，閱讀訓(xùn)練和測試數(shù)據(jù)集。每個示例都是評論及其標(biāo)簽：1 表示“正面”，0 表示“負(fù)面”。

#@save
def read_imdb(data_dir, is_train):
  """Read the IMDb review dataset text sequences and labels."""
  data, labels = [], []
  for label in ('pos', 'neg'):
    folder_name = os.path.join(data_dir, 'train' if is_train else 'test',
                  label)
    for file in os.listdir(folder_name):
      with open(os.path.join(folder_name, file), 'rb') as f:
        review = f.read().decode('utf-8').replace('n', '')
        data.append(review)
        labels.append(1 if label == 'pos' else 0)
  return data, labels

train_data = read_imdb(data_dir, is_train=True)
print('# trainings:', len(train_data[0]))
for x, y in zip(train_data[0][:3], train_data[1][:3]):
  print('label:', y, 'review:', x[:60])

# trainings: 25000
label: 1 review: Henry Hathaway was daring, as well as enthusiastic, for his
label: 1 review: An unassuming, subtle and lean film, "The Man in the White S
label: 1 review: Eddie Murphy really made me laugh my ass off on this HBO sta

#@save
def read_imdb(data_dir, is_train):
  """Read the IMDb review dataset text sequences and labels."""
  data, labels = [], []
  for label in ('pos', 'neg'):
    folder_name = os.path.join(data_dir, 'train' if is_train else 'test',
                  label)
    for file in os.listdir(folder_name):
      with open(os.path.join(folder_name, file), 'rb') as f:
        review = f.read().decode('utf-8').replace('n', '')
        data.append(review)
        labels.append(1 if label == 'pos' else 0)
  return data, labels

train_data = read_imdb(data_dir, is_train=True)
print('# trainings:', len(train_data[0]))
for x, y in zip(train_data[0][:3], train_data[1][:3]):
  print('label:', y, 'review:', x[:60])

# trainings: 25000
label: 1 review: Henry Hathaway was daring, as well as enthusiastic, for his
label: 1 review: An unassuming, subtle and lean film, "The Man in the White S
label: 1 review: Eddie Murphy really made me laugh my ass off on this HBO sta

16.1.2。預(yù)處理數(shù)據(jù)集

將每個單詞視為一個標(biāo)記并過濾掉出現(xiàn)次數(shù)少于 5 次的單詞，我們從訓(xùn)練數(shù)據(jù)集中創(chuàng)建了一個詞匯表。

train_tokens = d2l.tokenize(train_data[0], token='word')
vocab = d2l.Vocab(train_tokens, min_freq=5, reserved_tokens=[''])

train_tokens = d2l.tokenize(train_data[0], token='word')
vocab = d2l.Vocab(train_tokens, min_freq=5, reserved_tokens=[''])

標(biāo)記化后，讓我們繪制以標(biāo)記為單位的評論長度直方圖。

d2l.set_figsize()
d2l.plt.xlabel('# tokens per review')
d2l.plt.ylabel('count')
d2l.plt.hist([len(line) for line in train_tokens], bins=range(0, 1000, 50));

d2l.set_figsize()
d2l.plt.xlabel('# tokens per review')
d2l.plt.ylabel('count')
d2l.plt.hist([len(line) for line in train_tokens], bins=range(0, 1000, 50));

正如我們所料，評論的長度各不相同。為了每次處理一小批此類評論，我們將每個評論的長度設(shè)置為 500，并進(jìn)行截斷和填充，這類似于第 10.5 節(jié)中機器翻譯數(shù)據(jù)集的預(yù)處理 步驟。

num_steps = 500 # sequence length
train_features = torch.tensor([d2l.truncate_pad(
  vocab[line], num_steps, vocab['']) for line in train_tokens])
print(train_features.shape)

torch.Size([25000, 500])

num_steps = 500 # sequence length
train_features = np.array([d2l.truncate_pad(
  vocab[line], num_steps, vocab['']) for line in train_tokens])
print(train_features.shape)

(25000, 500)

16.1.3。創(chuàng)建數(shù)據(jù)迭代器

現(xiàn)在我們可以創(chuàng)建數(shù)據(jù)迭代器。在每次迭代中，返回一小批示例。

train_iter = d2l.load_array((train_features, torch.tensor(train_data[1])), 64)

for X, y in train_iter:
  print('X:', X.shape, ', y:', y.shape)
  break
print('# batches:', len(train_iter))

X: torch.Size([64, 500]) , y: torch.Size([64])
# batches: 391

train_iter = d2l.load_array((train_features, train_data[1]), 64)

for X, y in train_iter:
  print('X:', X.shape, ', y:', y.shape)
  break
print('# batches:', len(train_iter))

X: (64, 500) , y: (64,)
# batches: 391

16.1.4。把它們放在一起

最后，我們將上述步驟包裝到函數(shù)中l(wèi)oad_data_imdb。它返回訓(xùn)練和測試數(shù)據(jù)迭代器以及 IMDb 評論數(shù)據(jù)集的詞匯表。

#@save
def load_data_imdb(batch_size, num_steps=500):
  """Return data iterators and the vocabulary of the IMDb review dataset."""
  data_dir = d2l.download_extract('aclImdb', 'aclImdb')
  train_data = read_imdb(data_dir, True)
  test_data = read_imdb(data_dir, False)
  train_tokens = d2l.tokenize(train_data[0], token='word')
  test_tokens = d2l.tokenize(test_data[0], token='word')
  vocab = d2l.Vocab(train_tokens, min_freq=5)
  train_features = torch.tensor([d2l.truncate_pad(
    vocab[line], num_steps, vocab['']) for line in train_tokens])
  test_features = torch.tensor([d2l.truncate_pad(
    vocab[line], num_steps, vocab['']) for line in test_tokens])
  train_iter = d2l.load_array((train_features, torch.tensor(train_data[1])),
                batch_size)
  test_iter = d2l.load_array((test_features, torch.tensor(test_data[1])),
                batch_size,
                is_train=False)
  return train_iter, test_iter, vocab

#@save
def load_data_imdb(batch_size, num_steps=500):
  """Return data iterators and the vocabulary of the IMDb review dataset."""
  data_dir = d2l.download_extract('aclImdb', 'aclImdb')
  train_data = read_imdb(data_dir, True)
  test_data = read_imdb(data_dir, False)
  train_tokens = d2l.tokenize(train_data[0], token='word')
  test_tokens = d2l.tokenize(test_data[0], token='word')
  vocab = d2l.Vocab(train_tokens, min_freq=5)
  train_features = np.array([d2l.truncate_pad(
    vocab[line], num_steps, vocab['']) for line in train_tokens])
  test_features = np.array([d2l.truncate_pad(
    vocab[line], num_steps, vocab['']) for line in test_tokens])
  train_iter = d2l.load_array((train_features, train_data[1]), batch_size)
  test_iter = d2l.load_array((test_features, test_data[1]), batch_size,
                is_train=False)
  return train_iter, test_iter, vocab

16.1.5。概括

情感分析研究人們在其生成的文本中的情感，這被認(rèn)為是將變長文本序列轉(zhuǎn)換為固定長度文本類別的文本分類問題。

預(yù)處理后，我們可以將斯坦福的大型電影評論數(shù)據(jù)集（IMDb 評論數(shù)據(jù)集）加載到帶有詞匯表的數(shù)據(jù)迭代器中。

16.1.6。練習(xí)

我們可以修改本節(jié)中的哪些超參數(shù)來加速訓(xùn)練情緒分析模型？

你能實現(xiàn)一個函數(shù)來將亞馬遜評論的數(shù)據(jù)集加載到數(shù)據(jù)迭代器和標(biāo)簽中以進(jìn)行情感分析嗎？