def find_winner(sentences, obj_a, obj_b, aspects):
"""
Finds the winner of two objects for given sentences and aspects. Returns a dictionary
containing all the information of the comparison (winner, sentences for each object, score for
each object and more).
sentences: Dictionary
dictionary containing sentences
obj_a: Argument
the first competing object
obj_b: Argument
the second competing object
aspects: List
list of Aspects
"""
if len(sentences) > 0:
max_sentscore = max(sentence.ES_score for sentence in sentences)
for sentence in sentences:
comp_result = what_is_better(sentence.text, obj_a, obj_b)
sentence.set_confidence(comp_result['marker_cnt'])
if comp_result['winner'] == obj_a: # objectA won the sentence
add_points(find_aspects(sentence.text, aspects), obj_a, sentence,
max_sentscore, score_function)
else: # objectB won the sentence
add_points(find_aspects(sentence.text, aspects), obj_b, sentence,
max_sentscore, score_function)
obj_a.sentences = prepare_sentence_list(obj_a.sentences)
obj_b.sentences = prepare_sentence_list(obj_b.sentences)
return build_final_dict(obj_a, obj_b, sentences)
def count_confindences(prepared_sentences, classification_results, aspects):
perfect = 0
excellent = 0
good = 0
medium = 0
ok = 0
for index, row in prepared_sentences.iterrows():
if len(find_aspects(row['sentence'], aspects)) == 0:
continue
label = classification_results['max'][index]
if label != 'NONE':
confidence = classification_results[label][index]
if confidence > 0.8:
perfect += 1
elif confidence > 0.7:
excellent += 1
elif confidence > 0.6:
good += 1
elif confidence > 0.5:
medium += 1
else:
ok += 1
excellent += perfect
good += excellent
medium += good
ok += medium
return (excellent, good, medium, ok)
def evaluate(sentences, prepared_sentences, classification_results, obj_a,
obj_b, aspects):
if len(sentences) > 0:
max_sentscore = max(sentence.ES_score for sentence in sentences)
counts = count_confindences(prepared_sentences, classification_results,
aspects)
threshold_sentences = find_threshold(counts, 5)
threshold_score = find_threshold(counts, 3)
for index, row in prepared_sentences.iterrows():
label = classification_results['max'][index]
if label == 'NONE':
continue
classification_confidence = classification_results[label][index]
sentence_text = row['sentence']
for s in sentences:
if s.text == sentence_text:
sentence = s
break
sentences.remove(sentence)
sentence.set_confidence(classification_confidence.item())
contained_aspects = find_aspects(sentence.text, aspects)
if (label == 'BETTER' and row['object_a'] == obj_a.name) or (
label == 'WORSE' and row['object_b'] == obj_a.name):
add_points(contained_aspects, obj_a, sentence, max_sentscore,
score_function, threshold_sentences, threshold_score)
else:
add_points(contained_aspects, obj_b, sentence, max_sentscore,
score_function, threshold_sentences, threshold_score)
with open('../config.json') as json_data_file:
config = json.load(json_data_file)
if config['use_heuristics']:
for aspect in aspects:
negation_dissolve_heuristic(obj_a, obj_b, aspect.name, aspects,
threshold_score)
negation_dissolve_heuristic(obj_b, obj_a, aspect.name, aspects,
threshold_score)
obj_a.sentences = prepare_sentence_list(obj_a.sentences)
obj_b.sentences = prepare_sentence_list(obj_b.sentences)
return build_final_dict(obj_a, obj_b, sentences)
def move_assignment(sentences_to_move, from_object, to_object, aspect, aspects,
threshold_score):
points_to_move = 0
points_for_multiple = 0
print('For', '\'' + aspect + '\'', 'there were moved',
len(sentences_to_move), 'sentences from', from_object.name, 'to',
to_object.name, '.')
print('----')
for sentence in sentences_to_move:
points = (
sentence.CAM_score
) / 10 if sentence.confidence < threshold_score else sentence.CAM_score
if len(find_aspects(sentence.text, aspects)) > 1:
points_for_multiple = points_for_multiple + points
else:
points_to_move = points_to_move + points
print('-' +
re.sub(' +', ' ', re.sub('[^a-zA-Z0-9 ]', ' ', sentence.text)))
from_object.sentences = [
sentence for sentence in from_object.sentences
if sentence not in sentences_to_move
]
from_object.points[aspect] = from_object.points[aspect] - points_to_move
from_object.totalPoints = from_object.totalPoints - \
(points_to_move + points_for_multiple)
to_object.sentences.extend(sentences_to_move)
to_object.add_points(aspect, points_to_move)
if points_for_multiple > 0:
from_object.points['multiple'] = from_object.points['multiple'] - \
points_for_multiple
to_object.add_points('multiple', points_for_multiple)
print((points_to_move /
(to_object.totalPoints + from_object.totalPoints)) * 100,
'% moved (total)')
print((points_to_move /
(to_object.points[aspect] + from_object.points[aspect])) * 100,
'% moved for', aspect)
print('----')