def main() -> None:
tripadvisor_train = LIBRARY_DATASETS / 'absa' / \
'tripadvisor_co_uk-travel_restaurant_reviews_sample_2000_train.csv'
parser = argparse.ArgumentParser(description='ABSA Train')
parser.add_argument('--rerank-model',
type=validate_existing_filepath,
default=None,
help='Path to rerank model .h5 file')
group = parser.add_mutually_exclusive_group()
group.add_argument('--data',
type=validate_existing_path,
default=tripadvisor_train,
help='Path to raw data (directory or txt/csv file)')
group.add_argument('--parsed-data',
type=validate_existing_directory,
default=None,
help='Path to parsed data directory')
args = parser.parse_args()
train = TrainSentiment(parse=not args.parsed_data,
rerank_model=args.rerank_model)
opinion_lex, aspect_lex = train.run(data=args.data,
parsed_data=args.parsed_data)
print('Aspect Lexicon: {}\n'.format(aspect_lex) + '=' * 40 + '\n')
print('Opinion Lexicon: {}'.format(opinion_lex))
def main() -> None:
lib_root = Path(path.realpath(__file__)).parent.parent.parent.parent
tripadvisor_train = (
lib_root / "datasets" / "absa" /
"tripadvisor_co_uk-travel_restaurant_reviews_sample_2000_train.csv")
parser = argparse.ArgumentParser(description="ABSA Train")
parser.add_argument(
"--rerank-model",
type=validate_existing_filepath,
default=None,
help="Path to rerank model .h5 file",
)
group = parser.add_mutually_exclusive_group()
group.add_argument(
"--data",
type=validate_existing_path,
default=tripadvisor_train,
help="Path to raw data (directory or txt/csv file)",
)
group.add_argument(
"--parsed-data",
type=validate_existing_directory,
default=None,
help="Path to parsed data directory",
)
args = parser.parse_args()
train = TrainSentiment(parse=not args.parsed_data,
rerank_model=args.rerank_model)
opinion_lex, aspect_lex = train.run(data=args.data,
parsed_data=args.parsed_data)
print(f"Done.\nAspect and Opinion lexicons written to {LEXICONS_OUT}.")
def train_file_callback(attr, old, new):
global train_data
SENTIMENT_OUT.mkdir(parents=True, exist_ok=True)
train = TrainSentiment(parse=True, rerank_model=None)
if len(train_src.data['file_contents']) == 1:
train_data = read_csv(train_src, index_cols=0)
file_name = train_src.data['file_name'][0]
raw_data_path = SENTIMENT_OUT / file_name
train_data.to_csv(raw_data_path, header=False)
print(f'Running_SentimentTraining on data...')
train.run(data=raw_data_path)
else:
f_contents = train_src.data['file_contents']
f_names = train_src.data['file_name']
raw_data_path = SENTIMENT_OUT / train_src.data['file_name'][
0].split('/')[0]
if not os.path.exists(raw_data_path):
os.makedirs(raw_data_path)
for f_content, f_name in zip(f_contents, f_names):
read_parsed_files(f_content, f_name)
print(f'Running_SentimentTraining on data...')
train.run(parsed_data=raw_data_path)
text_status.value = "Lexicon extraction completed"
with io.open(AcquireTerms.acquired_aspect_terms_path, "r") as fp:
aspect_data_csv = fp.read()
file_data = base64.b64encode(str.encode(aspect_data_csv))
file_data = file_data.decode("utf-8")
asp_src.data = {
'file_contents': [file_data],
'file_name': ['nameFile.csv']
}
out_path = RerankTerms.out_dir / 'generated_opinion_lex_reranked.csv'
with io.open(out_path, "r") as fp:
opinion_data_csv = fp.read()
file_data = base64.b64encode(str.encode(opinion_data_csv))
file_data = file_data.decode("utf-8")
op_src.data = {
'file_contents': [file_data],
'file_name': ['nameFile.csv']
}
clothing_train = os.path.join(args.data_folder,
'clothing_data/clothing_absa_train_small.csv')
if args.large == 'yes':
print(f'Using large dataset: clothing_data/clothing_absa_train.csv')
clothing_train = os.path.join(args.data_folder,
'clothing_data/clothing_absa_train.csv')
else:
print(f'Using small dataset: clothing_data/clothing_absa_train_small.csv')
clothing_train = os.path.join(
args.data_folder, 'clothing_data/clothing_absa_train_small.csv')
os.makedirs('outputs', exist_ok=True)
train = TrainSentiment(asp_thresh=args.asp_thresh,
op_thresh=args.op_thresh,
max_iter=args.max_iter)
opinion_lex, aspect_lex = train.run(data=clothing_train, out_dir='./outputs')
#Copy lexicons to outputs folder
asp_lex = shutil.copy(LEXICONS_OUT / 'generated_aspect_lex.csv', './outputs')
op_lex = shutil.copy(LEXICONS_OUT / 'generated_opinion_lex_reranked.csv',
'./outputs')
# Evaluation
# Although ABSA is an unsupervised method it can be metriced with a small sample of labeled data
def doc2IO(doc):
"""
Converts ABSA doc to IO span format for evaluation