-
Notifications
You must be signed in to change notification settings - Fork 0
/
build_graph.py
311 lines (251 loc) · 11 KB
/
build_graph.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
import re
import numpy as np
from collections import defaultdict
from scipy import sparse
from utlis import *
import torch
from torch_geometric.data import Data
from torch_geometric.utils import dense_to_sparse
from nltk.stem.porter import *
from nltk.stem.porter import *
import math
import pickle as pkl
import nltk
from nltk.corpus import stopwords
import heapq
from operator import itemgetter
from utlis import *
import pickle
with open(f"{args.dataset}_TF_IDF.pkl", "rb") as f:
doc_tfidf_values = pickle.load(f)
nltk.download('stopwords')
stop_words = set(stopwords.words('english'))
stemmer = PorterStemmer()
def few_shot(mask):
mask = mask.numpy()
true_index = np.where(mask==True)[0]
np.random.shuffle(true_index)
print("The shots are:", args.shots)
few_mask = true_index[0:args.shots] #random pick few answers
new_mask = np.zeros(mask.shape)
new_mask[few_mask]= True
return torch.tensor(new_mask)
def remove_high_freq_word(word_freq, stop_words):
high_fre_word =[]
k_keys_sorted = heapq.nlargest(args.remove_high_freq, word_freq.items(), key=itemgetter(1))
for tu in k_keys_sorted:
high_fre_word.append(tu[0])
print("high_fre_word", high_fre_word)
stop_words = stop_words.union(set(high_fre_word))
return stop_words
def Jaccard_Similarity(doc1, doc2):
# List the unique words in a document
words_doc1 = set(doc1.lower().split())
words_doc2 = set(doc2.lower().split())
# Find the intersection of words list of doc1 & doc2
intersection = words_doc1.intersection(words_doc2)
# Find the union of words list of doc1 & doc2
union = words_doc1.union(words_doc2)
# Calculate Jaccard similarity score
# using length of intersection set divided by length of union set
return float(len(intersection)) / len(union)
def clean(str,stop_words):
str_list_words = str.split()
clean_list_words = []
for word in str_list_words:
word = re.sub(r"[^a-zA-Z ]+", '', word)#.lower()
if word not in stop_words:
word = stemmer.stem(word)
clean_list_words.append(word)
return clean_list_words
def clean_remove(str, word_freq,stop_words):
str_list_words = str.split()
clean_list_words = []
#high_stop_words = remove_high_freq_word(word_freq, stop_words)
for word in str_list_words:
word = re.sub(r"[^a-zA-Z ]+", '', word)#.lower()
if word not in stop_words:
word = stemmer.stem(word)
if word_freq[word] >= args.low_fre_word:
clean_list_words.append(word)
return clean_list_words
def build_set(triplet_list):
words_raw = set()
entities = set()
relation = set()
word_freq = {}
for triplet in triplet_list:
#print("triplet", triplet)
if triplet != None:
try:
entity1, rel, entity2 = triplet
except:
entity1, rel, *entity2 = triplet
entity2 = ' '.join(entity2)
if entity1 != None:
entity1_words = clean(entity1,stop_words)
words_raw.update(entity1_words)
for word in entity1_words:
if word in word_freq:
word_freq[word] += 1
else:
word_freq[word] = 1
entities.add(entity1)
if entity2 != None:
entity2_words = clean(entity2,stop_words)
words_raw.update(entity2_words)
for word in entity2_words:
if word in word_freq:
word_freq[word] += 1
else:
word_freq[word] = 1
entities.add(entity2)
if rel != None:
rel_words = clean(rel,stop_words)
words_raw.update(rel_words)
for word in rel_words:
if word in word_freq:
word_freq[word] += 1
else:
word_freq[word] = 1
relation.add(rel)
#print(word_freq)
if args.remove_high_freq > 0:
high_stop_words = remove_high_freq_word(word_freq, stop_words)
else:
high_stop_words = stop_words
words = set()
for word in sorted(words_raw):
if word_freq[word] >= args.low_fre_word:
words.add(word)
# Step 2: form a vocab based on these words and save the vocab to a txt file
with open(f'./data/corpus/{args.dataset}_vocab.txt', "w") as f:
for word in sorted(words):
f.write(word + "\n")
return words, word_freq, entities, relation, high_stop_words
def build_adj_matrix(abstract_list, doc_list, triplet_list):
# Step 1: split the triplets into words
words, word_freq, entities, relation, stop_words = build_set(triplet_list)
# Step 3: using all the vocab to build an adjacent matrix
vocab_size = len(words)
adj_matrix = np.zeros((vocab_size, vocab_size), dtype=int)
word_to_index = {word: i for i, word in enumerate(sorted(words))}
#with open(f'./ablation_study/{args.dataset}.pickle', 'wb') as handle:
# pickle.dump(word_to_index, handle, protocol=pickle.HIGHEST_PROTOCOL)
entity_to_index = {entity: i + len(words) for i, entity in enumerate(sorted(entities))}
# Step 4: connect the entities and relations with edges
edges = defaultdict(int)
for triplet in triplet_list:
if triplet != None:
try:
entity1, rel, entity2 = triplet
except:
entity1, rel, *entity2 = triplet
entity2 = ' '.join(entity2)
if entity1 != None and entity2 != None:
clean_entity1 = clean_remove(entity1, word_freq,stop_words)
clean_entity2 = clean_remove(entity2, word_freq,stop_words)
for word1 in clean_entity1:
for word2 in clean_entity2:
if word1 != word2:
word1_index = word_to_index[word1]
word2_index = word_to_index[word2]
edge1 = (word1_index, word2_index)
edge2 = (word2_index, word1_index)
edges[edge1] += 10
edges[edge2] += 10
if entity1 != None and rel != None:
clean_entity1 = clean_remove(entity1, word_freq,stop_words)
clean_rel = clean_remove(rel , word_freq,stop_words)
for word1 in clean_entity1:
for word2 in clean_rel:
if word1 != word2:
word1_index = word_to_index[word1]
word2_index = word_to_index[word2]
edge1 = (word1_index, word2_index)
edge2 = (word2_index, word1_index)
edges[edge1] += 1
edges[edge2] += 1
if entity2 != None and rel != None:
clean_entity2 = clean_remove(entity2, word_freq,stop_words)
clean_rel = clean_remove(rel, word_freq,stop_words)
for word1 in clean_entity2:
for word2 in clean_rel:
if word1 != word2:
word1_index = word_to_index[word1]
word2_index = word_to_index[word2]
edge1 = (word1_index, word2_index)
edge2 = (word2_index, word1_index)
edges[edge1] += 1
edges[edge2] += 1
# Step 5: connect words within the same entity with edges
for entity in entities:
clean_entity_words = clean_remove(entity, word_freq,stop_words)
for i, word1 in enumerate(clean_entity_words):
for j, word2 in enumerate(clean_entity_words):
if i != j and word1!= word2:
word1_index = word_to_index[word1]
word2_index = word_to_index[word2]
edge1 = (word1_index, word2_index)
edge2 = (word2_index, word1_index)
edges[edge1] += 1 #/(len(clean_entity_words))
edges[edge2] += 1 #/(len(clean_entity_words))
# Add edges to the adjacency matrix
for edge, count in edges.items():
row, col = edge
adj_matrix[row, col] = count
pooling_index = torch.zeros((vocab_size, len(doc_list)))
for i, doc in enumerate(doc_list):
doc = doc.replace(';',', ')
doc = clean(doc,stop_words)
for word in doc:
try:
pooling_index[word_to_index[word]][i] = doc_tfidf_values[i].get(word, 0)
#pooling_index[word_to_index[word]][i] =+1
except:
pass
pooling_index = pooling_index/(pooling_index.sum(dim=0)+0.01)
return torch.tensor(adj_matrix), pooling_index, word_to_index
def build_feature(doc_list, triplet_list, word_to_index):
words,_, _, _, _ = build_set(triplet_list)
num_node = len(words) #+ len(doc_list)
emb_size = args.emb_size
emb_matrix = torch.rand((num_node,emb_size))
feature_matrix = torch.eye(num_node)
return feature_matrix, emb_matrix
def build_label(dataset):
label_set = set()
label_set = set()
label_list = []
f = open('data/' + dataset + '.txt', 'r')
lines = f.readlines()
for line in lines:
temp = line.split("\t")
label_set.add(temp[2].strip())
label_list.append(temp[2].strip())
f.close()
unique_labels = np.unique(label_list)
label_tensor = torch.tensor(np.searchsorted(unique_labels, label_list))
return label_tensor
def build_graph(dataset):
abs_doc_list, KG_doc_list, KG_list = read_KG(args.dataset)
adj, pool_index, word_to_index = build_adj_matrix(abs_doc_list, KG_doc_list, KG_list)
edge_index = dense_to_sparse(adj)[0]
edge_attr = dense_to_sparse(adj)[1].float()
feature_matrix, emb_matrix = build_feature(KG_doc_list, KG_list, word_to_index)
doc_train_mask, doc_val_mask, doc_test_mask = split_shuffle(dataset)
if args.usewhole:
print("Using the whole graph")
train_mask = doc_train_mask.bool()
else:
train_mask = few_shot(doc_train_mask).bool()
print("training dataset samples:", train_mask.sum())
val_mask = doc_val_mask.bool()
test_mask = doc_test_mask.bool()
doc_label = build_label(args.dataset)
y_label = doc_label.long()
print(train_mask)
print(val_mask)
print(test_mask)
doc_graph = Data(x=feature_matrix, edge_index=edge_index, edge_attr=edge_attr, y = y_label, train_mask=train_mask, val_mask=val_mask, test_mask=test_mask, pool_index= pool_index, emb_matrix=emb_matrix)
return doc_graph