-
Notifications
You must be signed in to change notification settings - Fork 16
/
document_similarity.py
331 lines (274 loc) · 9.1 KB
/
document_similarity.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
import pickle
import nltk
import re
import pandas as pd
import numpy as np
from nltk.tokenize import word_tokenize
from nltk.stem import WordNetLemmatizer
from nltk.stem.porter import PorterStemmer
from nltk.corpus import stopwords
from sklearn.pipeline import Pipeline, FeatureUnion
from sklearn.feature_extraction.text import CountVectorizer, TfidfTransformer
from sklearn.metrics.pairwise import euclidean_distances
from gensim.models.word2vec import Word2Vec
import constants as const
from gensim.models.doc2vec import Doc2Vec
from gensim.models.doc2vec import TaggedDocument
from sentence_transformers import SentenceTransformer
nltk.download(['punkt', 'wordnet', 'averaged_perceptron_tagger', 'stopwords'])
def tokenize(text):
"""Tokenize the text
Parameters
----------
text: String
The message to be tokenized
Returns
-------
List
List with the clean tokens
"""
text = text.lower()
text = re.sub("[^a-zA-Z0-9]", " ", text)
tokens = word_tokenize(text)
tokens = [w for w in tokens if w not in stopwords.words(const.ENGLISH)]
lemmatizer = WordNetLemmatizer()
stemmer = PorterStemmer()
clean_tokens_list = []
for tok in tokens:
lemmatizer_tok = lemmatizer.lemmatize(tok).strip()
clean_tok = stemmer.stem(lemmatizer_tok)
clean_tokens_list.append(clean_tok)
return clean_tokens_list
def build_model():
"""Build the model
Returns
-------
sklearn.pipeline.Pipeline
The model
"""
pipeline = Pipeline([
(const.FEATURES, FeatureUnion([
(const.TEXT_PIPELINE, Pipeline([
(const.VECT, CountVectorizer(tokenizer=tokenize)),
(const.TFIDF, TfidfTransformer())
]))
]))])
return pipeline
def get_avg_document_vector(model, df, year):
"""Get the a vector representation of a document using word2vec
Parameters
----------
model: Word2Vec
Trained Word2Vec model
df: pandas DataFrame
Pandas DataFrame with a columns with the tokens
year: int
The target year
Returns
-------
Tuple
The vector representation of the document, the number os words that are not at the model
vocabulary
"""
word_vecs = []
count = 0
for word in df[const.TOKENIZED].loc[year]:
try:
vector = model[word]
word_vecs.append(vector)
except KeyError:
count += 1
pass
vector_avg = np.mean(word_vecs, axis=0)
return vector_avg, count
def get_letters_df(letters_dict_pickle):
"""Get the letters Pandas Dataframe
Parameters
----------
letters_dict_pickle: string
Path to the dict with the letters text
Returns
-------
Pandas DataFrame
Pandas DataFrame with a columns with the tokens
"""
with open(letters_dict_pickle, 'rb') as handle:
letters_dict = pickle.load(handle)
letters_df = pd.DataFrame(letters_dict, index=[const.LETTER_TEXT]).T
letters_df[const.TOKENIZED] = letters_df[const.LETTER_TEXT].apply(tokenize)
return letters_df
def get_most_similar_docs(pairwise_similarities, letter_year, distance_method, transformers=False, initial_year=1977):
"""Get the most similar letters to a target one
Parameters
----------
pairwise_similarities: np.array
Numpy array of the pairwise similarities
letter_year: int
The target letter year
distance_method: string
Euclidean or cosine
transformers: boolean
True if you are calling from transformers or False otherwise
initial_year: int
The initial letter year
Returns
-------
List
List with the letter year sorted descending by similarity
"""
letter_i = letter_year - initial_year
if distance_method == const.COSINE:
if transformers:
similarity_index = np.array(np.argsort(-pairwise_similarities[letter_i]))
else:
similarity_index = np.array(np.argsort(-pairwise_similarities[letter_i].todense()))[0]
else:
similarity_index = np.argsort(pairwise_similarities[letter_i])
similar_docs_sorted = []
for index in similarity_index:
if index == letter_i:
continue
similar_docs_sorted.append(index + initial_year)
return similar_docs_sorted
def get_pipe_vector(letters_df):
"""Get the tfidf vector
Parameters
----------
letters_df: pandas DataFrame
The pandas Dataframe with text from the letters
Returns
-------
Np.array
The tfidf vector representation of the text
"""
pipeline = build_model()
pipeline.fit(letters_df[const.LETTER_TEXT])
vectors = pipeline.transform(letters_df[const.LETTER_TEXT])
return vectors
def get_tfidf(letters_df, year, n, distance):
"""Get the tfidf most similar years
Parameters
----------
letters_df: pandas DataFrame
The pandas Dataframe with text from the letters
year: int
The target letter year
n: int
The number of letters to return
distance: string
Euclidean or cosine
Returns
-------
List
List with the letter year sorted descending by similarity
"""
vectors = get_pipe_vector(letters_df)
if distance == const.COSINE:
pairwise_dis = vectors @ vectors.T
else:
pairwise_dis = euclidean_distances(vectors)
return get_most_similar_docs(pairwise_dis, year, distance)[:n]
def get_most_similar_docs_docs2vec(letter_year, model, corpus, n, initial_year=1977):
"""Get the docs2vec most similar years
Parameters
----------
letter_year: int
The target letter year
model: docs2vec
The trained Docs2vec model
corpus: List
TaggedDocument list
n: int
The number of letters to return
initial_year: int
The initial letter year
Returns
-------
List
List with the letter year sorted descending by similarity
"""
doc_id = letter_year - initial_year
inferred_vector = model.infer_vector(corpus[doc_id].words)
sims = model.docvecs.most_similar([inferred_vector], topn=len(model.docvecs))
sims = [index + initial_year for index, _ in sims]
return sims[1:n + 1]
def get_doc2vec(letters_df, year, n):
"""Get the doc2vec most similar years
Parameters
----------
letters_df: pandas DataFrame
The pandas Dataframe with text from the letters
year: int
The target letter year
n: int
The number of letters to return
Returns
-------
List
List with the letter year sorted descending by similarity
"""
EPOCHS = 40
doc2_model = Doc2Vec(min_count=2)
corpus = [TaggedDocument(tokens, [i]) for i, tokens in enumerate(list(letters_df[const.TOKENIZED]))]
doc2_model.build_vocab(corpus)
doc2_model.train(corpus, total_examples=doc2_model.corpus_count, epochs=EPOCHS)
return get_most_similar_docs_docs2vec(year, doc2_model, corpus, n)
def get_word2vec(letters_df, year, n, distance):
"""Get the word2vec most similar years
Parameters
----------
letters_df: pandas DataFrame
The pandas Dataframe with text from the letters
year: int
The target letter year
n: int
The number of letters to return
distance: string
Euclidean or cosine
Returns
-------
List
List with the letter year sorted descending by similarity
"""
model = Word2Vec(letters_df[const.TOKENIZED])
target, _ = get_avg_document_vector(model, letters_df, year)
distances = []
for y in list(letters_df.index):
if y != year:
vector_year, _ = get_avg_document_vector(model, letters_df, y)
if distance == const.COSINE:
distances.append(target @ vector_year.T / np.linalg.norm(target) / np.linalg.norm(vector_year))
else:
distances.append(np.linalg.norm(target - vector_year))
distances = np.array(distances)
if distance == const.COSINE:
return letters_df.index[(-distances).argsort()][:n]
else:
return letters_df.index[distances.argsort()][:n]
def get_transformers(pre_trained_model, letters_df, year, n, distance):
"""Get the word2vec most similar years
Parameters
----------
pre_trained_model: string
The name of the pre trained transform
letters_df: pandas DataFrame
The pandas Dataframe with text from the letters
year: int
The target letter year
n: int
The number of letters to return
distance: string
Euclidean or cosine
Returns
-------
List
List with the letter year sorted descending by similarity
"""
model = SentenceTransformer(pre_trained_model)
embeddings = model.encode(letters_df[const.TOKENIZED].values)
if distance == const.COSINE:
pairwise = embeddings @ embeddings.T / np.linalg.norm(embeddings) / np.linalg.norm(embeddings)
return get_most_similar_docs(pairwise, year, const.COSINE, transformers=True)[:n]
else:
euclidean = euclidean_distances(embeddings)
return get_most_similar_docs(euclidean, year, const.EUCLIDEAN)[:n]