-
Notifications
You must be signed in to change notification settings - Fork 666
/
14_nlp_class.py
345 lines (274 loc) · 10.6 KB
/
14_nlp_class.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
'''
CLASS: Natural Language Processing
Adapted from: https://github.com/charlieg/A-Smattering-of-NLP-in-Python
What is NLP?
- Using computers to process (analyze, understand, generate) natural human languages
Why NLP?
- Most knowledge created by humans is unstructured text
- Need some way to make sense of it
- Enables quantitative analysis of text data
Why NLTK?
- High-quality, reusable NLP functionality
'''
import nltk
nltk.download('all')
'''
Tokenization
What: Separate text into units such as sentences or words
Why: Gives structure to previously unstructured text
Notes: Relatively easy with English language text, not easy with some languages
'''
# "corpus" = collection of documents
# "corpora" = plural form of corpus
import requests
from bs4 import BeautifulSoup
r = requests.get("http://en.wikipedia.org/wiki/Data_science")
b = BeautifulSoup(r.text)
paragraphs = b.find("body").findAll("p")
text = ""
for paragraph in paragraphs:
text += paragraph.text + " "
# Data Science corpus
text[:500]
# tokenize into sentences
sentences = [sent for sent in nltk.sent_tokenize(text)]
sentences[:10]
# tokenize into words
tokens = [word for word in nltk.word_tokenize(text)]
tokens[:100]
# only keep tokens that start with a letter (using regular expressions)
import re
clean_tokens = [token for token in tokens if re.search('^[a-zA-Z]+', token)]
clean_tokens[:100]
# count the tokens
from collections import Counter
c = Counter(clean_tokens)
c.most_common(25) # mixed case
sorted(c.items())[:25] # counts similar words separately
for item in sorted(c.items())[:25]:
print item[0], item[1]
'''
Stemming
What: Reduce a word to its base/stem form
Why: Often makes sense to treat multiple word forms the same way
Notes: Uses a "simple" and fast rule-based approach
Output can be undesirable for irregular words
Stemmed words are usually not shown to users (used for analysis/indexing)
Some search engines treat words with the same stem as synonyms
'''
from nltk.stem.snowball import SnowballStemmer
stemmer = SnowballStemmer('english')
# example stemming
stemmer.stem('charge')
stemmer.stem('charging')
stemmer.stem('charged')
# stem the tokens
stemmed_tokens = [stemmer.stem(t) for t in clean_tokens]
# count the stemmed tokens
c = Counter(stemmed_tokens)
c.most_common(25) # all lowercase
sorted(c.items())[:25] # some are strange
'''
Lemmatization
What: Derive the canonical form ('lemma') of a word
Why: Can be better than stemming, reduces words to a 'normal' form.
Notes: Uses a dictionary-based approach (slower than stemming)
'''
lemmatizer = nltk.WordNetLemmatizer()
# compare stemmer to lemmatizer
stemmer.stem('dogs')
lemmatizer.lemmatize('dogs')
stemmer.stem('wolves') # Beter for information retrieval and search
lemmatizer.lemmatize('wolves') # Better for text analysis
stemmer.stem('is')
lemmatizer.lemmatize('is')
lemmatizer.lemmatize('is',pos='v')
'''
Part of Speech Tagging
What: Determine the part of speech of a word
Why: This can inform other methods and models such as Named Entity Recognition
Notes: http://www.ling.upenn.edu/courses/Fall_2003/ling001/penn_treebank_pos.html
'''
temp_sent = 'Sinan and Kevin are great teachers!'
# pos_tag takes a tokenize sentence
nltk.pos_tag(nltk.word_tokenize(temp_sent))
'''
Stopword Removal
What: Remove common words that will likely appear in any text
Why: They don't tell you much about your text
'''
# most of top 25 stemmed tokens are "worthless"
c.most_common(25)
# view the list of stopwords
stopwords = nltk.corpus.stopwords.words('english')
sorted(stopwords)
# stem the stopwords
stemmed_stops = [stemmer.stem(t) for t in stopwords]
# remove stopwords from stemmed tokens
stemmed_tokens_no_stop = [stemmer.stem(t) for t in stemmed_tokens if t not in stemmed_stops]
c = Counter(stemmed_tokens_no_stop)
most_common_stemmed = c.most_common(25)
# remove stopwords from cleaned tokens
clean_tokens_no_stop = [t for t in clean_tokens if t not in stopwords]
c = Counter(clean_tokens_no_stop)
most_common_not_stemmed = c.most_common(25)
# Compare the most common results for stemmed words and non stemmed words
for i in range(25):
text_list = most_common_stemmed[i][0] + ' ' + str(most_common_stemmed[i][1]) + ' '*25
text_list = text_list[0:30]
text_list += most_common_not_stemmed[i][0] + ' ' + str(most_common_not_stemmed[i][1])
print text_list
'''
Named Entity Recognition
What: Automatically extract the names of people, places, organizations, etc.
Why: Can help you to identify "important" words
Notes: Training NER classifier requires a lot of annotated training data
Should be trained on data relevant to your task
Stanford NER classifier is the "gold standard"
'''
def extract_entities(text):
entities = []
# tokenize into sentences
for sentence in nltk.sent_tokenize(text):
# tokenize sentences into words
# add part-of-speech tags
# use NLTK's NER classifier
chunks = nltk.ne_chunk(nltk.pos_tag(nltk.word_tokenize(sentence)))
# parse the results
entities.extend([chunk for chunk in chunks if hasattr(chunk, 'label')])
return entities
for entity in extract_entities('Kevin and Sinan are instructors for General Assembly in Washington, D.C.'):
print '[' + entity.label() + '] ' + ' '.join(c[0] for c in entity.leaves())
'''
Term Frequency - Inverse Document Frequency (TF-IDF)
What: Computes "relative frequency" that a word appears in a document
compared to its frequency across all documents
Why: More useful than "term frequency" for identifying "important" words in
each document (high frequency in that document, low frequency in
other documents)
Notes: Used for search engine scoring, text summarization, document clustering
'''
sample = ['Bob likes sports', 'Bob hates sports', 'Bob likes likes trees']
from sklearn.feature_extraction.text import CountVectorizer
vect = CountVectorizer()
vect.fit_transform(sample).toarray()
vect.get_feature_names()
from sklearn.feature_extraction.text import TfidfVectorizer
tfidf = TfidfVectorizer()
tfidf.fit_transform(sample).toarray()
tfidf.get_feature_names()
'''
LDA - Latent Dirichlet Allocation
What: Way of automatically discovering topics from sentences
Why: Much quicker than manually creating and identifying topic clusters
'''
import lda
import numpy as np
# Instantiate a count vectorizer with two additional parameters
vect = CountVectorizer(stop_words='english', ngram_range=[1,3])
sentences_train = vect.fit_transform(sentences)
# Instantiate an LDA model
model = lda.LDA(n_topics=10, n_iter=500)
model.fit(sentences_train) # Fit the model
n_top_words = 10
topic_word = model.topic_word_
for i, topic_dist in enumerate(topic_word):
topic_words = np.array(vect.get_feature_names())[np.argsort(topic_dist)][:-n_top_words:-1]
print('Topic {}: {}'.format(i, ', '.join(topic_words)))
'''
EXAMPLE: Automatically summarize a document
'''
# corpus of 2000 movie reviews
from nltk.corpus import movie_reviews
reviews = [movie_reviews.raw(filename) for filename in movie_reviews.fileids()]
# create document-term matrix
tfidf = TfidfVectorizer(stop_words='english')
dtm = tfidf.fit_transform(reviews)
features = tfidf.get_feature_names()
# find the most and least "interesting" sentences in a randomly selected review
def summarize():
# choose a random movie review
review_id = np.random.randint(0, len(reviews))
review_text = reviews[review_id]
# we are going to score each sentence in the review for "interesting-ness"
sent_scores = []
# tokenize document into sentences
for sentence in nltk.sent_tokenize(review_text):
# exclude short sentences
if len(sentence) > 6:
score = 0
token_count = 0
# tokenize sentence into words
tokens = nltk.word_tokenize(sentence)
# compute sentence "score" by summing TFIDF for each word
for token in tokens:
if token in features:
score += dtm[review_id, features.index(token)]
token_count += 1
# divide score by number of tokens
sent_scores.append((score / float(token_count + 1), sentence))
# lowest scoring sentences
print '\nLOWEST:\n'
for sent_score in sorted(sent_scores)[:3]:
print sent_score[1]
# highest scoring sentences
print '\nHIGHEST:\n'
for sent_score in sorted(sent_scores, reverse=True)[:3]:
print sent_score[1]
# try it out!
summarize()
'''
TextBlob Demo: "Simplified Text Processing"
Installation: pip install textblob
'''
from textblob import TextBlob, Word
# identify words and noun phrases
blob = TextBlob('Kevin and Sinan are instructors for General Assembly in Washington, D.C.')
blob.words
blob.noun_phrases
# sentiment analysis
blob = TextBlob('I hate this horrible movie. This movie is not very good.')
blob.sentences
blob.sentiment.polarity
[sent.sentiment.polarity for sent in blob.sentences]
# sentiment subjectivity
TextBlob("I am a cool person").sentiment.subjectivity # Pretty subjective
TextBlob("I am a person").sentiment.subjectivity # Pretty objective
# different scores for essentially the same sentence
print TextBlob('Kevin and Sinan are instructors for General Assembly in Washington, D.C.').sentiment.subjectivity
print TextBlob('Kevin and Sinan are instructors in Washington, D.C.').sentiment.subjectivity
# singularize and pluralize
blob = TextBlob('Put away the dishes.')
[word.singularize() for word in blob.words]
[word.pluralize() for word in blob.words]
# spelling correction
blob = TextBlob('15 minuets late')
blob.correct()
# spellcheck
Word('parot').spellcheck()
# definitions
Word('bank').define()
Word('bank').define('v')
# translation and language identification
blob = TextBlob('Welcome to the classroom.')
blob.translate(to='es')
blob = TextBlob('Hola amigos')
blob.detect_language()
'''
Data Science Toolkit Sentiment
Provides many different APIs for converting and getting information
We'll use the text2sentiment API.
'''
# Import the necessary modules
import requests
import json
# Sample sentences
sentences = ['I love Sinan!', 'I hate Sinan!', 'I feel nothing about Sinan!']
# API endpoint (i.e.the URL they ask you to send your text to)
url = 'http://www.datasciencetoolkit.org/text2sentiment/'
# Loop through the sentences
for sentence in sentences:
payload = {'text': sentence} # The sentence we want the sentiment of
headers = {'content-type': 'application/json'} # The type of data you are sending
r = requests.post(url, data=json.dumps(payload), headers=headers) # Send the data
print sentence, json.loads(r.text)['score'] # Print the results