-
Notifications
You must be signed in to change notification settings - Fork 2
/
interpretation.py
361 lines (298 loc) · 13.4 KB
/
interpretation.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
346
347
348
import dill
import argparse
import plotly.graph_objects as go
import numpy as np
import unidecode
import os
from collections import Counter
import pandas as pd
from sklearn.feature_extraction.text import TfidfVectorizer
from collections import defaultdict
from nltk.corpus import stopwords
def get_stopwords(lang):
if lang == 'slo':
sw = []
with open('resources/stopwords.txt', 'r', encoding='utf8') as f:
for line in f:
sw.append(unidecode.unidecode(line.strip()))
return set(sw)
elif lang == 'en':
return set(stopwords.words('english'))
def get_clusters_sent(target, threshold_size_cluster, labels, sentences, id2sents, corpus_slices, docs_folder):
labels = dill.load(open(labels, 'rb'))
sentences = dill.load(open(sentences, 'rb'))
id2sents = dill.load(open(id2sents, 'rb'))
cluster_to_sentence = defaultdict(lambda: defaultdict(list))
for cs in corpus_slices:
for label, sents in zip(labels[target][cs], sentences[target][cs]):
for sent in sents:
sent_id = int(str(corpus_slices.index(cs) + 1) + str(sent))
sent = id2sents[sent_id]
cluster_to_sentence[label][cs].append(sent)
counts = {cs: Counter(labels[target][cs]) for cs in corpus_slices}
all_labels = []
for slice, c in counts.items():
slice_labels = [x[0] for x in c.items()]
all_labels.extend(slice_labels)
all_labels = set(all_labels)
all_counts = []
for l in all_labels:
all_count = 0
for slice in corpus_slices:
count = counts[slice][l]
all_count += count
all_counts.append((l, all_count))
sorted_counts = sorted(all_counts, key=lambda x: x[1], reverse=True)
sentences = []
lemmas = []
metas = []
labels = []
categs = []
for label, count in sorted_counts:
if count > threshold_size_cluster:
for cs in corpus_slices:
for (sent, lemma, meta) in cluster_to_sentence[label][cs]:
sent_clean = sent.strip()
lemma_clean = lemma.replace('_<ner>', '').strip()
time_token = '<' + str(cs) + '>'
sent_clean = sent_clean.replace(time_token, '').strip()
lemma_clean = lemma_clean.replace(time_token, '').strip()
if sent_clean not in set(sentences):
sentences.append(sent_clean)
lemmas.append(lemma_clean)
labels.append(label)
categs.append(cs)
metas.append(meta)
#print(sent_clean)
else:
print("Cluster", label, "is too small - deleted!")
all_metas = []
if metas:
columns = list(metas[0].keys())
meta_values = [list(x.values()) for x in metas]
for i in range(len(metas[0])):
meta_value = [x[i] for x in meta_values]
all_metas.append(meta_value)
else:
columns = []
columns = columns + ['slice', 'cluster_label', 'sentence', 'lemmatized_sent']
output = all_metas + [categs, labels, sentences, lemmas]
sent_df = pd.DataFrame(list(zip(*output)),
columns=columns)
sent_df.to_csv(os.path.join(docs_folder, target + '_sentences.tsv'), encoding='utf-8', index=False, sep='\t')
return sent_df
def output_distrib(data, word, keyword_clusters, image_folder):
distrib = data.groupby(['slice', "cluster_label"]).size().reset_index(name="count")
pivot_distrib = distrib.pivot(index='slice', columns='cluster_label', values='count')
pivot_distrib_norm = pivot_distrib.div(pivot_distrib.sum(axis=1), axis=0)
pivot_distrib_norm = pivot_distrib_norm.fillna(0)
first_column = pivot_distrib_norm.columns[0]
order = list('Slice ' + x for x in pivot_distrib_norm[first_column].keys())
columns = []
final_data = []
for i in keyword_clusters:
name = "Cluster " + str(i) + ": " + ", ".join(keyword_clusters[i][:7])
distrib = np.array(list(pivot_distrib_norm[i].fillna(0).array))
final_data.append((name, distrib))
final_data = sorted(final_data, reverse=True, key=lambda x:sum(x[1]))
if len(final_data) <= 10:
for name, distrib in final_data:
columns.append(go.Bar(name=name, x=order, y=distrib))
else:
for name, distrib in final_data[:9]:
columns.append(go.Bar(name=name, x=order, y=distrib))
other_data = final_data[9:]
other = None
print(other_data)
for name, distrib in other_data:
if other is None:
other = distrib
print(distrib, other)
else:
other += distrib
print(distrib, other)
print('Other: ', other)
columns.append(go.Bar(name='Other', x=order, y=other))
fig = go.Figure(data=columns)
fig.update_layout(
margin=dict(l=20, r=20, t=20, b=20),
width=1200,
height=800,
barmode='stack',
title='',
xaxis_title="Slice",
yaxis_title="Distribution",
legend_title="",
font=dict(
size=14,
color="Black"
),
legend = dict(
yanchor="top",
y=1.4,
xanchor="left",
x=0.3
)
)
#fig.show()
fig.write_image(os.path.join(image_folder, f"{word}.png"))
return pivot_distrib_norm
def extract_topn_from_vector(feature_names, sorted_items, topn):
"""get the feature names and tf-idf score of top n items"""
# use only topn items from vector
sorted_items = sorted_items[:topn]
score_vals = []
feature_vals = []
for idx, score in sorted_items:
fname = feature_names[idx]
# keep track of feature name and its corresponding score
score_vals.append(round(score, 3))
feature_vals.append(feature_names[idx])
# create a tuples of feature,score
# results = zip(feature_vals,score_vals)
results = {}
for idx in range(len(feature_vals)):
results[feature_vals[idx]] = score_vals[idx]
return results
def extract_keywords(target_word, word_clustered_data, max_df, topn, lang):
sw = get_stopwords(lang)
# get groups of sentences for each cluster
l_sent_clust_dict = defaultdict(list)
sent_clust_dict = defaultdict(list)
for i, row in word_clustered_data.iterrows():
sent = " ".join(row['sentence'].split())
lemmatized_sent = " ".join(row['lemmatized_sent'].split())
l_sent_clust_dict[row['cluster_label']].append((sent, lemmatized_sent))
for label, data in l_sent_clust_dict.items():
original_sents = "\t".join([x[0] for x in data])
lemmas = "\t".join([x[1] for x in data])
sent_clust_dict[label] = (original_sents, lemmas)
labels = []
lemmatized_clusters = []
for label, (sents, lemmatized_sents) in sent_clust_dict.items():
labels.append(label)
lemmatized_clusters.append(lemmatized_sents)
tfidf_transformer = TfidfVectorizer(smooth_idf=True, use_idf=True, ngram_range=(1,3), max_df=max_df, max_features=10000)
tfidf_transformer.fit(lemmatized_clusters)
feature_names = tfidf_transformer.get_feature_names()
keyword_clusters = {}
for label, lemmatized_cluster in zip(labels, lemmatized_clusters):
# generate tf-idf
tf_idf_vector = tfidf_transformer.transform([lemmatized_cluster])
# sort the tf-idf vectors by descending order of scores
tuples = zip(tf_idf_vector.tocoo().col, tf_idf_vector.tocoo().data)
sorted_items = sorted(tuples, key=lambda x: (x[1], x[0]), reverse=True)
# extract only the top n
keywords = extract_topn_from_vector(feature_names, sorted_items, topn*10)
keywords = sorted(keywords.items(), key=lambda x: x[1], reverse=True)
scores = {x[0]:x[1] for x in keywords}
already_in = set()
filtered_keywords = []
for kw, score in keywords:
if len(kw.split()) == 1:
for k, s in scores.items():
if kw in k and len(k.split()) > 1:
if score > s:
already_in.add(k)
else:
already_in.add(kw)
if len(kw.split()) == 2:
for k, s in scores.items():
if kw in k and len(k.split()) > 2:
if score > s:
already_in.add(k)
else:
already_in.add(kw)
if kw not in already_in and kw != target_word:
filtered_keywords.append(kw)
already_in.add(kw)
keyword_clusters[label] = filtered_keywords[:topn*10]
final_keywords = {}
all_data = []
for c, keywords in keyword_clusters.items():
sents = sent_clust_dict[c][0].split('\t')
lemmas = sent_clust_dict[c][1].split('\t')
all_sents = " ".join(sents)
all_lemmas = " ".join(lemmas)
set_lemmatized_sents = set(lemmas)
filtered_keywords = []
for kw in keywords:
stop = 0
for word in kw.split():
if word in sw:
stop += 1
if stop / float(len(kw.split())) < 0.5:
num_appearances = 0
for sent in set_lemmatized_sents:
if kw in sent:
num_appearances += 1
if num_appearances > 1:
if len(kw) > 2:
if kw + ' ' + target_word in all_lemmas:
kw = kw + ' ' + target_word
elif target_word + ' ' + kw in all_lemmas:
kw = target_word + ' ' + kw
filtered_keywords.append(kw)
if len(filtered_keywords) == 0:
filtered_keywords.append('other')
final_keywords[c] = filtered_keywords[:topn]
all_data.append((c, ";".join(filtered_keywords[:50]), all_sents))
return final_keywords
def full_analysis(word, labels, sentences, id2sent, corpus_slices, image_folder, docs_folder, max_df=0.8, topn=15, threshold_size_cluster=10, lang='slo'):
clusters_sents_df = get_clusters_sent(word, threshold_size_cluster, labels, sentences, id2sent, corpus_slices, docs_folder)
keyword_clusters = extract_keywords(word, clusters_sents_df, topn=topn, max_df=max_df, lang=lang)
for k in keyword_clusters:
keywords = keyword_clusters[k]
output_distrib(clusters_sents_df, word, keyword_clusters, image_folder)
return keyword_clusters
def loadData(labels_path, sentences_path, id2sents_path):
labels = dill.load(open(labels_path, 'rb'))
sentences = dill.load(open(sentences_path, 'rb'))
id2sents = dill.load(open(id2sents_path, 'rb'))
return labels, sentences, id2sents
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Interpret changes')
parser.add_argument('--target_words',
default="test",
type=str,
help='Target words to analyse, separated by comma')
parser.add_argument("--lang",
default='slo',
type=str,
help="Language of the corpus, currently only Slovenian ('slo') and English ('en') are supported")
parser.add_argument("--input_dir",
default="output",
type=str,
help="Folder containing data generated by the script 'measure_semantic_shift.py'")
parser.add_argument("--results_dir",
default="results",
type=str,
help="Path to final results")
parser.add_argument('--max_df', type=float, default=0.8,
help='Words that appear in more than that percentage of clusters will not be used as keywords.')
parser.add_argument('--cluster_size_threshold', type=int, default=10,
help='Clusters smaller than a threshold will be deleted.')
parser.add_argument('--num_keywords', type=int, default=10, help='Number of keywords per cluster.')
args = parser.parse_args()
args = parser.parse_args()
image_folder = os.path.join(args.results_dir, "images")
docs_folder = os.path.join(args.results_dir, "docs")
target_words = args.target_words.split(',')
if not os.path.exists(args.results_dir):
os.makedirs(args.results_dir)
if not os.path.exists(image_folder):
os.makedirs(image_folder)
if not os.path.exists(docs_folder):
os.makedirs(docs_folder)
labels = os.path.join(args.input_dir, "kmeans_5_labels.pkl")
sentences = os.path.join(args.input_dir, "sents.pkl")
id2sent = os.path.join(args.input_dir, "id2sents.pkl")
corpus_slices = os.path.join(args.input_dir, "corpus_slices.pkl")
corpus_slices = dill.load(open(corpus_slices, 'rb'))
for word in target_words:
print('Generating results for word:', word)
keyword_clusters = full_analysis(word, labels, sentences, id2sent, corpus_slices, image_folder,
docs_folder, max_df=args.max_df, topn=args.num_keywords,
threshold_size_cluster=args.cluster_size_threshold, lang=args.lang)
print("Results written to folder:", args.results_dir)
print('--------------------------------------------')