def annotate_sentences(model, sents):
cared_sents = []
embeddings = word2vec(sents) # If use USC
for i in range(len(sents)):
sentence = sents[i]
if finding_by_heuristics(sentence):
cared_sents.append(sentence)
continue
# Word2Vec feature
feature1 = embeddings[i].tolist() # If use USC
# feature1 = word2vec(sentence) # Use law2vec
# Basic features: sentence length, number of periods, percent of characters that are capitalized
feature2 = basic_features(sentence)
# Fraction of POS feature
feature3 = frac_of_pos(sentence)
# Fraction of years and numbers feature
feature4 = frac_of_years_and_numbers(sentence)
# Cue words feature
feature5 = cue_words(sentence)
features = np.array(feature1 + feature2 + feature3 + feature4 +
feature5).reshape(1, -1)
type = model.predict(features)
if type[0] == 1:
cared_sents.append(sentence)
return cared_sents
def combine_features(cases_names):
types = [
'Citation', 'LegalRule', 'LegalPolicy', 'PolicyBasedReasoning',
'ConclusionOfLaw', 'EvidenceBasedFinding', 'EvidenceBasedReasoning',
'Evidence', 'Procedure', 'Header'
]
type_set = set(types)
x = []
y = []
for case_name in cases_names:
annotated_sentences = read_files([case_name])
if not os.path.exists('../annotated_casetext/' + case_name[:-4] +
'.pickle'):
pure_sentences = [
annotated_sentences[i][0]
for i in range(len(annotated_sentences))
]
embeddings = word2vec(pure_sentences)
# Save embeddings to speed up
with open('../annotated_casetext/' + case_name[:-4] + '.pickle',
'wb') as handle:
pickle.dump(embeddings, handle, protocol=2)
else:
with open('../annotated_casetext/' + case_name[:-4] + '.pickle',
'rb') as handle:
embeddings = pickle.load(handle)
for i in list(range(len(annotated_sentences))):
sentence, sent_type = annotated_sentences[i]
label = 0
if sent_type == 'EvidenceBasedFinding':
label = 1
elif sent_type != 'Evidence':
continue
y.append(label)
# Word2Vec feature
feature1 = embeddings[i].tolist()
# Basic features: sentence length, number of periods, percent of characters that are capitalized
feature2 = basic_features(sentence)
# Fraction of POS feature
feature3 = frac_of_pos(sentence)
# Fraction of years and numbers feature
feature4 = frac_of_years_and_numbers(sentence)
# Cue words feature
feature5 = cue_words(sentence)
features = feature1 + feature2 + feature3 + feature4 + feature5
x.append(np.array(features))
x = np.array(x)
y = np.array(y)
return x, y
def combine_features(cases_names):
x = []
y = []
for case_name in cases_names:
annotated_sentences = read_files([case_name], annotated_dir)
if not os.path.exists(annotated_dir + case_name[:-4] + '.pickle'):
pure_sentences = [
annotated_sentences[i][0]
for i in range(len(annotated_sentences))
]
try:
embeddings = word2vec(pure_sentences)
except:
print("USC failed", case_name)
continue
# Save embeddings to speed up
with open(annotated_dir + case_name[:-4] + '.pickle',
'wb') as handle:
pickle.dump(embeddings, handle, protocol=2)
else:
with open(annotated_dir + case_name[:-4] + '.pickle',
'rb') as handle:
embeddings = pickle.load(handle)
for i in range(len(annotated_sentences)):
sentence, sent_type = annotated_sentences[i]
if sent_type in finding_types or sent_type in fact_types:
label = 1
# elif sent_type in fact_types:
# label = 2
else:
label = 0
y.append(label)
# Word2Vec feature
feature1 = embeddings[i].tolist()
# Basic features: sentence length, number of periods, percent of characters that are capitalized
feature2 = basic_features(sentence)
# Fraction of POS feature
feature3 = frac_of_pos(sentence)
# Fraction of years and numbers feature
feature4 = frac_of_years_and_numbers(sentence)
# Cue words feature
feature5 = cue_words(sentence)
features = feature1 + feature2 + feature3 + feature4 + feature5
x.append(np.array(features))
x = np.array(x)
y = np.array(y)
return x, y