r"""
Add the difference part of target in ABSA task
==========================================================
"""
import random
import string
from .absa_transformation import ABSATransformation
__all__ = ['AddDiff']
[docs]class AddDiff(ABSATransformation):
r"""
Add the difference part of aspect to the end of original sentence.
The difference part is extracted from the training set of SemEval2014
or user's customization.
Example::
Original sentence: "BEST spicy tuna roll, great asian salad.
（Target: spicy tuna roll)"
Transformed sentence: "BEST spicy tuna roll, great asian salad,
but this small place is packed,
on a cold day, the seating by the entrance way can be pretty drafty and
bad service."
"""
def __init__(
self,
language="eng"):
super().__init__()
if language != "eng":
raise ValueError(f"Language {language} is not available.")
self.language = language
self.tokenize = self.processor.tokenize
self.untokenize = self.processor.inverse_tokenize
def __repr__(self):
return "AddDiff"
def _transform(self, sample, n=1, field='sentence',
extra_text=None, **kwargs):
r"""
Transform data sample to a list of Sample.
:param ~textflint.input_layer.component.sample.ABSAsample sample: input ABSAsample
:param int n: the number of transformation, in ABSA-specific transformations n=1
:param str field:field name
:param dict extra_text: extra text will be added to the original sentence
:return list: list of transformed ABSAsample
"""
trans_samples = []
self.trans_term_list = sample.term_list
self.terms = sample.terms
self.term_list = sample.term_list
self.sentence = sample.sentence.text
if sample.trans_id is None:
self.trans_id = [idx for idx in self.terms]
else:
self.trans_id = [sample.trans_id]
all_term = [self.term_list[idx]['term'] for idx in self.term_list]
for term_id in self.trans_id:
trans_sample = sample.clone(sample)
polarity = self.term_list[term_id]['polarity']
if polarity == 'positive':
add_text = extra_text['negative']
elif polarity == 'negative':
add_text = extra_text['positive']
else:
add_text = extra_text['neutral']
add_sentence = self._get_add_sentence(add_text, all_term)
trans_sentence = self._concatenate_sentence(add_sentence)
trans_sample.update_sentence(trans_sentence)
trans_sample.trans_id = term_id
trans_samples.append(trans_sample)
return trans_samples
def _get_add_sentence(self, add_text, all_term):
r"""
Get the sentence that owns different polarity compared with
the aspect. Choose 1~3 sentences randomly from add_text and put
them together.
:param list add_text: extra text
:param list all_term: all aspect term
:return list: extra sentence that need to be added to original sentence
"""
sentence = self.sentence
punctuation = '.'
if sentence[-1] == string.punctuation:
punctuation = sentence[-1]
while True:
add_num = random.randint(1, 3)
rand_num = random.sample(range(len(add_text)), 3)
rand_terms = [add_text[i][0] for i in rand_num]
rand_sentences = [self.tokenize(add_text[i][1][0])
for i in rand_num]
for idx, ran_sentence in enumerate(rand_sentences):
if rand_terms[idx] in all_term:
break
if ran_sentence[-1] in string.punctuation:
rand_sentences[idx] = ran_sentence[:-1]
if add_num == 3:
add_sentence = rand_sentences[0] + [','] + rand_sentences[1] + [
'and'] + rand_sentences[2] + [punctuation]
elif add_num == 2:
add_sentence = rand_sentences[0] + ['and'] + rand_sentences[1] \
+ [punctuation]
else:
add_sentence = rand_sentences[0] + [punctuation]
break
return add_sentence
def _concatenate_sentence(self, add_sentence):
r"""
Concatenate the extra part to original sentence.
:param list add_sentence: extra sentence that need to be
added to original sentence
:return list: transformed sentence
"""
sentence = self.sentence[:-1]
opi_tag = self.get_postag(add_sentence, 0, 1)
if opi_tag[0] != 'CONJ':
trans_sentence = self.untokenize(add_sentence)
if 'but' in sentence or 'although' in sentence:
trans_sentence = sentence + "; " + trans_sentence
else:
trans_sentence = sentence + ", but " + trans_sentence
else:
trans_sentence = self.untokenize(add_sentence)
trans_sentence = sentence + ". " + trans_sentence[
0].upper() + trans_sentence[1:]
return trans_sentence