-
Notifications
You must be signed in to change notification settings - Fork 2
/
data_preprocess_semeval.py
181 lines (146 loc) · 5.99 KB
/
data_preprocess_semeval.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
'''
Biaffine Dependency parser from AllenNLP
'''
import argparse
import json
import os
import re
import sys
from allennlp.predictors.predictor import Predictor
from lxml import etree
from nltk.tokenize import TreebankWordTokenizer
from tqdm import tqdm
MODELS_DIR = '/data1/yangyy/pretrained-models'
model_path = os.path.join(
MODELS_DIR, "biaffine-dependency-parser-ptb-2018.08.23.tar.gz")
def parse_args():
parser = argparse.ArgumentParser()
# Required parameters
parser.add_argument('--model_path', type=str, default=model_path,
help='Path to biaffine dependency parser.')
parser.add_argument('--data_path', type=str, default='/data1/SHENWZH/ABSA_online/data/semeval14',
help='Directory of where semeval14 or twiiter data held.')
return parser.parse_args()
def xml2txt(file_path):
'''
Read the original xml file of semeval data and extract the text that have aspect terms.
Store them in txt file.
'''
output = file_path.replace('.xml', '_text.txt')
sent_list = []
with open(file_path, 'rb') as f:
raw = f.read()
root = etree.fromstring(raw)
for sentence in root:
sent = sentence.find('text').text
terms = sentence.find('aspectTerms')
if terms is None:
continue
if terms:
sent_list.append(sent)
with open(output, 'w') as f:
for s in sent_list:
f.write(s+'\n')
print('processed', len(sent_list), 'of', file_path)
def text2docs(file_path, predictor):
'''
Annotate the sentences from extracted txt file using AllenNLP's predictor.
'''
with open(file_path, 'r') as f:
sentences = f.readlines()
docs = []
print('Predicting dependency information...')
for i in tqdm(range(len(sentences))):
docs.append(predictor.predict(sentence=sentences[i]))
return docs
def dependencies2format(doc): # doc.sentences[i]
'''
Format annotation: sentence of keys
- tokens
- tags
- predicted_dependencies
- predicted_heads
- dependencies
'''
sentence = {}
sentence['tokens'] = doc['words']
sentence['tags'] = doc['pos']
# sentence['energy'] = doc['energy']
predicted_dependencies = doc['predicted_dependencies']
predicted_heads = doc['predicted_heads']
sentence['predicted_dependencies'] = doc['predicted_dependencies']
sentence['predicted_heads'] = doc['predicted_heads']
sentence['dependencies'] = []
for idx, item in enumerate(predicted_dependencies):
dep_tag = item
frm = predicted_heads[idx]
to = idx + 1
sentence['dependencies'].append([dep_tag, frm, to])
return sentence
def get_dependencies(file_path, predictor):
docs = text2docs(file_path, predictor)
sentences = [dependencies2format(doc) for doc in docs]
return sentences
def syntaxInfo2json(sentences, origin_file):
json_data = []
tk = TreebankWordTokenizer()
mismatch_counter = 0
idx = 0
with open(origin_file, 'rb') as fopen:
raw = fopen.read()
root = etree.fromstring(raw)
for sentence in root:
example = dict()
example["sentence"] = sentence.find('text').text
# for RAN
terms = sentence.find('aspectTerms')
if terms is None:
continue
example['tokens'] = sentences[idx]['tokens']
example['tags'] = sentences[idx]['tags']
example['predicted_dependencies'] = sentences[idx]['predicted_dependencies']
example['predicted_heads'] = sentences[idx]['predicted_heads']
example['dependencies'] = sentences[idx]['dependencies']
# example['energy'] = sentences[idx]['energy']
example["aspect_sentiment"] = []
example['from_to'] = [] #left and right offset of the target word
for c in terms:
if c.attrib['polarity'] == 'conflict':
continue
target = c.attrib['term']
example["aspect_sentiment"].append((target, c.attrib['polarity']))
# index in strings, we want index in tokens
left_index = int(c.attrib['from'])
right_index = int(c.attrib['to'])
left_word_offset = len(tk.tokenize(example['sentence'][:left_index]))
to_word_offset = len(tk.tokenize(example['sentence'][:right_index]))
example['from_to'].append((left_word_offset,to_word_offset))
if len(example['aspect_sentiment'])==0:
idx += 1
continue
json_data.append(example)
idx+=1
extended_filename = origin_file.replace('.xml', '_biaffine_depparsed.json')
with open(extended_filename, 'w') as f:
json.dump(json_data, f)
print('done', len(json_data))
print(idx)
def main():
args = parse_args()
predictor = Predictor.from_path(args.model_path)
data = [('Restaurants_Train_v2.xml', 'Restaurants_Test_Gold.xml'),
('Laptop_Train_v2.xml', 'Laptops_Test_Gold.xml')]
for train_file, test_file in data:
# xml -> txt
xml2txt(os.path.join(args.data_path, train_file))
xml2txt(os.path.join(args.data_path, test_file))
# txt -> json
train_sentences = get_dependencies(
os.path.join(args.data_path, train_file.replace('.xml', '_text.txt')), predictor)
test_sentences = get_dependencies(os.path.join(
args.data_path, test_file.replace('.xml', '_text.txt')), predictor)
print(len(train_sentences), len(test_sentences))
syntaxInfo2json(train_sentences, os.path.join(args.data_path, train_file))
syntaxInfo2json(test_sentences, os.path.join(args.data_path, test_file))
if __name__ == "__main__":
main()