Tutorial: Simple Text Classification with Python and TextBlob
Yesterday, TextBlob 0.6.0 was released (changelog), which introduces Naive Bayes classification. This tutorial shows how to use TextBlob to create your own text classification systems.
The tutorial assumes that you have TextBlob >= 0.6.0 and nltk >= 2.0 TextBlob >= 8.0 installed. If you don't yet have TextBlob or need to upgrade, run:
$ pip install -U textblob nltk
If this is your first time installing TextBlob, you may have to download the necessary NLTK corpora. This can be done with one command:
$ >python -m textblob.download_corpora
Part 1: A Tweet Sentiment Analyzer (Simple classification)
Our first classifier will be a simple sentiment analyzer trained on a small dataset of fake tweets.
To begin, we'll import the textblob.classifiers and create some training and test data.
from textblob.classifiers import NaiveBayesClassifier
train = [
('I love this sandwich.', 'pos'),
('This is an amazing place!', 'pos'),
('I feel very good about these beers.', 'pos'),
('This is my best work.', 'pos'),
("What an awesome view", 'pos'),
('I do not like this restaurant', 'neg'),
('I am tired of this stuff.', 'neg'),
("I can't deal with this", 'neg'),
('He is my sworn enemy!', 'neg'),
('My boss is horrible.', 'neg')
]
test = [
('The beer was good.', 'pos'),
('I do not enjoy my job', 'neg'),
("I ain't feeling dandy today.", 'neg'),
("I feel amazing!", 'pos'),
('Gary is a friend of mine.', 'pos'),
("I can't believe I'm doing this.", 'neg')
]
We create a new classifier by passing training data into the constructor for a NaiveBayesClassifier.
cl = NaiveBayesClassifier(train)
We can now classify arbitrary text using the NaiveBayesClassifier.classify(text) method.
cl.classify("Their burgers are amazing") # "pos"
cl.classify("I don't like their pizza.") # "neg"
Another way to classify strings of text is to use TextBlob objects.
You can pass classifiers into the constructor of a TextBlob.
from textblob import TextBlob
blob = TextBlob("The beer was amazing. "
"But the hangover was horrible. My boss was not happy.",
classifier=cl)
You can then call the classify() method on the blob.
blob.classify() # "neg"
You can also take advantage of TextBlob's sentence tokenization and classify each sentence indvidually.
for sentence in blob.sentences:
print(sentence)
print(sentence.classify())
# "pos", "neg", "neg"
Let's check the accuracy on the test set.
cl.accuracy(test)
# 0.83
We can also find the most informative features:
cl.show_informative_features(5)
# Most Informative Features
# contains(my) = True neg : pos = 1.7 : 1.0
# contains(an) = False neg : pos = 1.6 : 1.0
# contains(my) = False pos : neg = 1.3 : 1.0
# contains(place) = False neg : pos = 1.2 : 1.0
# contains(of) = False pos : neg = 1.2 : 1.0
This indicates that tweets containing the word "my" but not containing the word "place" tend to be negative.
Here is the full script:
Part 2: Adding More Data from NLTK
We can improve our classifier by adding more training and test data. Here we'll add data from the movie review corpus which was downloaded with NLTK.
import random
from nltk.corpus import movie_reviews
reviews = [(list(movie_reviews.words(fileid)), category)
for category in movie_reviews.categories()
for fileid in movie_reviews.fileids(category)]
new_train, new_test = reviews[0:100], reviews[101:200]
Let's see what one of these documents looks like.
print(new_train[0])
This outputs:
(['kolya', 'is', 'one', 'of', 'the', 'richest', 'films', 'i', "'", 've', 'seen', 'in', 'some', 'time'
, '.', 'zdenek', 'sverak', 'plays', 'a', 'confirmed', 'old', 'bachelor', '(', 'who', "'", 's', 'likel
y', 'to', 'remain', 'so', ')', ',', 'who', 'finds', 'his', 'life', 'as', 'a', 'czech', 'cellist', 'in
creasingly', 'impacted', 'by', 'the', 'five', '-', 'year', 'old', 'boy', 'that', 'he', "'", 's', 'tak
ing', 'care', 'of', '.', 'though', 'it', 'ends', 'rather', 'abruptly', '--', 'and', 'i', "'", 'm', 'w
hining', ',', "'", 'cause', 'i', 'wanted', 'to', 'spend', 'more', 'time', 'with', 'these', 'character
s', '--', 'the', 'acting', ',', 'writing', ',', 'and', 'production', 'values', 'are', 'as', 'high', '
as', ',', 'if', 'not', 'higher', 'than', ',', 'comparable', 'american', 'dramas', '.', 'this', 'fathe
r', '-', 'and', '-', 'son', 'delight', '--', 'sverak', 'also', 'wrote', 'the', 'script', ',', 'while'
, 'his', 'son', ',', 'jan', ',', 'directed', '--', 'won', 'a', 'golden', 'globe', 'for', 'best', 'for
eign', 'language', 'film', 'and', ',', 'a', 'couple', 'days', 'after', 'i', 'saw', 'it', ',', 'walked
', 'away', 'an', 'oscar', '.', 'in', 'czech', 'and', 'russian', ',', 'with', 'english', 'subtitles',
'.'], 'pos')
Notice that unlike the data in Part 1, the text comes as a list of words instead of a single string. TextBlob is smart about this; it will treat both forms of data as expected.
We can now update our classifier with the new training data using the update(new_data) method, as well as test it using the larger test dataset.
cl.update(new_train)
accuracy = cl.accuracy(test + new_test)
Here's the full, updated script:
Part 3: Language Detector (Custom Feature Extraction)
An important aspect that I haven't yet mentioned is how features are being extracted from the text.
For a given document and training set train, TextBlob's default behavior is to compute which words in train are present in document. For example, the sentence "It's just a flesh wound." might have features contains(flesh): True, contains(wound): True, and contains(knight): False.
Of course, this simple feature extractor may not be appropriate for all problems. Here we'll create a custom feature extractor for a language detector.
Here's the training and test data.
train = [
("amor", "spanish"),
("perro", "spanish"),
("playa", "spanish"),
("sal", "spanish"),
("oceano", "spanish"),
("love", "english"),
("dog", "english"),
("beach", "english"),
("salt", "english"),
("ocean", "english")
]
test = [
("ropa", "spanish"),
("comprar", "spanish"),
("camisa", "spanish"),
("agua", "spanish"),
("telefono", "spanish"),
("clothes", "english"),
("buy", "english"),
("shirt", "english"),
("water", "english"),
("telephone", "english")
]
A feature extractor is simply a function that takes an argument text (the text to extract features from) and returns a dictionary of features.
Let's create a very simple extractor that uses the last letter of a given word as its only feature.
def extractor(word):
feats = {}
last_letter = word[-1]
feats["last_letter({0})".format(last_letter)] = True
return feats
print(extractor("python")) # {'last_letter(n)': True}
We can pass this feature extractor as the second argument to the constructor of a NaiveBayesClassifier.
lang_detector = NaiveBayesClassifier(train, feature_extractor=extractor)
And again, compute accuracy and informative features.
lang_detector.accuracy(test) # 0.7
lang_detector.show_informative_features(5)
# Most Informative Features
# last_letter(o) = None englis : spanis = 1.6 : 1.0
# last_letter(l) = None englis : spanis = 1.2 : 1.0
# last_letter(n) = None spanis : englis = 1.2 : 1.0
# last_letter(h) = None spanis : englis = 1.2 : 1.0
# last_letter(e) = None spanis : englis = 1.2 : 1.0
Not surprisingly, words that do not end with the letter "o" tend to be English.
Full script:
Conclusion
TextBlob makes it easy to create your own custom text classifiers. Always remember that there is no free lunch in machine learning, and every problem requires extensive experimentation. Happy experimenting.
Further Reading
- TextBlob documentation
- Naive Bayes Classification on Wikipedia
- The NLTK Book, Ch. 6: Learning to Classify Text
Edits
January 12, 2016: Update imports in gists. Thanks Mark Menkhus.
October 26, 2014: Update imports for TextBlob>=0.8.0. Thanks Justyna Chromik.
February 22, 2014: Update command to download corpora. Thanks Jason.
Please send comments by email. I welcome your feedback, advice, and criticism.