Tokenization:
Tokenization means breaking down text into smaller units, such as words, phrases, or sentences.
Tokenization is an important concept in Natural Language Processing (NLP) that involves breaking down the text into smaller, more manageable units called tokens. This process is necessary because computers understand and process text in a very different way than humans do. Breaking down the text into individual tokens allows machines to analyze and interpret text more effectively. Many different models can be used for tokenization, including rule-based, statistical, and deep-learning models. Rule-based models use pre-defined rules to identify tokens, while statistical models use probabilities to determine the most likely tokens. Deep learning models use neural networks to learn patterns in text and identify tokens based on those patterns.
Examples
Some popular tokenization methods include:
- Whitespace tokenization is the simplest and most commonly used form of tokenization. It splits the text whenever it finds whitespace characters. The split() function can easily implement in many languages. But it also has limitations because spaces are not a good means for splitting sentences. It will consider “model” and “model.” as two different tokens, which may not be useful in some cases.
- Punctuation-based tokenization is one of the traditional methods used in natural language processing to split the text into smaller units called tokens. As the name suggests, this method involves breaking up text based on the presence of punctuation marks such as commas, periods, and semicolons. In this method, the text is split into individual words based on the position of punctuation marks. For example, for the sentence “Ulife is the best AI service provider for innovative, modern, and customer-first companies,” we would end up with the following tokens: “Ulife,” “is,” “the,” “best,” “AI,” “service,” “provider,” “for,” “innovative,” “,” (comma), “modern,” “,” (comma), “and,” “customer-first,” “companies,” “.” (period). However, it has some limitations. For instance, this method can’t handle words that contain punctuation marks, such as “don’t” or “isn’t.” Additionally, some languages, such as Chinese or Japanese, do not use punctuation marks the same way as Western languages, making this method less effective for these languages.
- Default/TreebankWordTokenizer is a popular word tokenizer in natural language processing based on the Penn Treebank corpus, a large annotated dataset of English text. This tokenizer is available as part of the Natural Language Toolkit (NLTK), a widely used library for NLP in Python. It can handle words with apostrophes, such as “don’t” or “can’t,” and also handles hyphenated words, such as “self-driving.”. It can also handle contractions, such as “won’t” or “shouldn’t,” by splitting them into their parts. However, It has its limitations and difficulties, such as text in languages other than English or text with non-standard punctuation or formatting. In such cases, other tokenization methods, such as regular expression or byte pair encoding tokenization, may be more suitable.
- …
We still have many possibilities with some tokenizer like ****TweetTokenizer (****as the name suggest, more accurate with tweets-based texts) and MWETokenizer (The multi-word expression tokenizer, which is specifically designed to handle multi-word expressions, which are fixed phrases or idiomatic expressions that consist of two or more words)
Usage in python
NLTK
Basic word tokenization:
The simplest way to use the word_tokenize function is to pass a string of text to it, and it will automatically split the text into individual words or tokens.
import nltkfrom nltk.tokenize import (TreebankWordTokenizer, word_tokenize, wordpunct_tokenize, TweetTokenizer, MWETokenizer)sentence = "Ulife is the best AI service provider for innovative, modern, and customer-first companies,"print(f'Output = {word_tokenize(sentence)}')Output:
Output = ['Ulife', 'is', 'the', 'best', 'AI', 'service', 'provider', 'for', 'innovative', ',', 'modern', ',', 'and', 'customer-first', 'companies', ',']Punctuation-based tokenizer
In that case, we must use wordpunct_tokenize ****function instead of word_tokenize
print(f'Output = {wordpunct_tokenize(sentence)}')Output:
Output = ['Ulife', 'is', 'the', 'best', 'AI', 'service', 'provider', 'for', 'innovative', ',', 'modern', ',', 'and', 'customer', '-', 'first', 'companies', ',']Treebank Word tokenizer, Tweet tokenizer and MWE tokenizer.
Here we must instantiate the classes we have imported previously:
# MWEtokenizer = MWETokenizer()tokenizer.add_mwe(('Martha', 'Jones'))print(f'MWE tokenization = {tokenizer.tokenize(word_tokenize(sentence))}')# TWeettokenizer = TweetTokenizer()print(f'Tweet-rules based tokenization = {tokenizer.tokenize(sentence)}')# TreebankWordTtokenizer = TreebankWordTokenizer()print(f'Default/Treebank tokenization = {tokenizer.tokenize(sentence)}')MWE tokenization = ['Ulife', 'is', 'the', 'best', 'AI', 'service', 'provider', 'for', 'innovative', ',', 'modern', ',', 'and', 'customer-first', 'companies', ',']Tweet-rules based tokenization = ['Ulife', 'is', 'the', 'best', 'AI', 'service', 'provider', 'for', 'innovative,', 'modern,', 'and', 'customer-first', 'companies', ',']Default/Treebank tokenization = ['Ulife', 'is', 'the', 'best', 'AI', 'service', 'provider', 'for', 'innovative', ',', 'modern', ',', 'and', 'customer-first', 'companies', ',']Tokenizing a paragraph into sentences:
word_tokenize can also be used to tokenize a paragraph of text into individual sentences.
from nltk.tokenize import sent_tokenize, word_tokenizetext = "This is the first sentence. This is the second sentence. And this is the third sentence."sentences = sent_tokenize(text)for sentence in sentences: words = word_tokenize(sentence) print(words)Output:
['This', 'is', 'the', 'first', 'sentence', '.']['This', 'is', 'the', 'second', 'sentence', '.']['And', 'this', 'is', 'the', 'third', 'sentence', '.']Tokenizing non-English text:
word_tokenize can also be used to tokenize non-English text. However, in this case, it is necessary to specify the language to ensure proper tokenization.
from nltk.tokenize import word_tokenizetext = "Das ist ein Beispieltext auf Deutsch."tokens = word_tokenize(text, language='german')print(tokens)Output:
['Das', 'ist', 'ein', 'Beispieltext', 'auf', 'Deutsch', '.']spaCy Tokenizer
spaCy is a popular open-source natural language processing library that provides many tools for processing and analyzing text data. One of the main components of spaCy is its tokenizer, which is responsible for breaking down a piece of text into individual tokens.
The spaCy tokenizer is based on the rule-based approach, where a set of rules is defined to determine how to split the input text into individual tokens. The tokenizer in spaCy is designed to work with various languages and handles a wide range of cases, including special characters, punctuation, and contractions.
One of the advantages of the spaCy tokenizer is its ability to handle complex cases, such as URLs, email addresses, and phone numbers. It also provides several built-in token attributes, such as lemma, part of speech, and named entities, which can be useful for downstream tasks like text classification and information extraction.
To use it you must first install and download the models:
$ pip install spacy$ python3 -m spacy download en_core_web_sm
Visit https://spacy.io/usage to choose the language and configuration that fit your needs.
import spacynlp = spacy.load("en_core_web_sm")doc = nlp("This is an example sentence.")# Iterate over the tokens in the documentfor token in doc: print(token.text)Output:
Thisisanexamplesentence.Tokenization with Keras
Keras is a popular open-source deep-learning library that provides a wide range of tools for building and training neural networks. It also includes several pre-processing functions, including text tokenization.
In Keras, the text tokenization function is provided by the Tokenizer class in the preprocessing.text module. The Tokenizer class provides a simple and efficient way to tokenize text data, with several options for customizing the tokenization process.
To use the Tokenizer class, you first need to create an instance of the class and fit it to your text data. This step builds the vocabulary of the tokenizer and assigns a unique integer index to each token in the vocabulary. Here is an example:
from keras.preprocessing.text import Tokenizer# Define some example text datatext_data = ["This is the first sentence.", "This is the second sentence."]# Create a tokenizer and fit it to the text datatokenizer = Tokenizer()tokenizer.fit_on_texts(text_data)# Print the vocabulary of the tokenizerprint(tokenizer.word_index)Output:
{'this': 1, 'is': 2, 'the': 3, 'first': 4, 'sentence': 5, 'second': 6}So, that is. I hope you found this article interesting and that it helps you implement AI into your product. Turn out it is exactly what we do at Ulife.ai. We help companies change their perspective with AI. We handle data collection, model evaluation, and product development. You can contact us here for more information.
