文字資料編碼

Deep Learning for Text with PyTorch

Shubham Jain

Data Scientist

文字編碼

Pytorch 處理流程

將文字轉成機器可讀的數值
便於分析與建模

從序列資料找洞見的示意圖

編碼技術

One-hot encoding： 將單字轉為唯一的數值表示
Bag-of-Words（BoW）： 擷取詞頻，不考慮順序
TF-IDF： 兼顧稀有度與重要性
Embedding： 將單字轉成向量，捕捉語意（第 2 章）

One-hot encoding

將每個單字對應到獨立向量
二元向量：
- 出現為 1
- 未出現為 0

['cat', 'dog', 'rabbit']
- 'cat' [1, 0, 0]
- 'dog' [0, 1, 0]
- 'rabbit' [0, 0, 1]

用 PyTorch 做 One-hot encoding

import torch
vocab = ['cat', 'dog', 'rabbit']

vocab_size = len(vocab)

one_hot_vectors = torch.eye(vocab_size)

one_hot_dict = {word: one_hot_vectors[i] for i, word in enumerate(vocab)}

print(one_hot_dict)

{'cat': tensor([1., 0., 0.]),
  'dog': tensor([0., 1., 0.]),
  'rabbit': tensor([0., 0., 1.])}

Bag-of-words

範例：「The cat sat on the mat」
Bag-of-words：
- {'the': 2, 'cat': 1, 'sat': 1, 'on': 1, 'mat': 1}

將每份文件視為無序的詞彙集合
著重於「頻率」，不看順序

CountVectorizer

from sklearn.feature_extraction.text import CountVectorizer

vectorizer = CountVectorizer()

corpus = ['This is the first document.', 'This document is the second document.', 
'And this is the third one.', 'Is this the first document?']

X = vectorizer.fit_transform(corpus)

print(X.toarray())

print(vectorizer.get_feature_names_out())

[[0 1 1 1 0 0 1 0 1]
 [0 2 0 1 0 1 1 0 1]
 [1 0 0 1 1 0 1 1 1]
 [0 1 1 1 0 0 1 0 1]]

 ['and' 'document' 'first' 'is' 'one' 'second' 'the' 'third' 'this']

TF-IDF

Term Frequency-Inverse Document Frequency
- 衡量單字在文件中的重要性
- 罕見詞得分較高
- 常見詞得分較低
- 強調具資訊性的詞

TfidfVectorizer

from sklearn.feature_extraction.text import TfidfVectorizer
vectorizer = TfidfVectorizer()

corpus = ['This is the first document.','This document is the second document.',
'And this is the third one.','Is this the first document?']

X = vectorizer.fit_transform(corpus)

print(X.toarray())

print(vectorizer.get_feature_names_out())

[[0.         0.         0.68091856 0.51785612 0.51785612 0.        ]
 [0.         0.         0.          0.51785612 0.51785612 0.68091856]
 [0.85151335 0.42575668 0.         0.32274454 0.32274454 0.        ]
 [0.         0.         0.68091856 0.51785612 0.51785612 0.        ]]

['and' 'document' 'first' 'is' 'one' 'second']

TfidfVectorizer

TFIDF 程式碼

編碼技術

技術：One-hot encoding、bag-of-words、TF-IDF

讓模型能理解並處理文字
選用一種技術以避免重複
還有更多方法可用

一起來練習吧！

Deep Learning for Text with PyTorch