def execute(self, sent):
features = sent2features_(sent)
# print("----------features_features--------------")
# print(features)
act_type = self.__predictor.predict([features])
# print("----------act_type--------------")
# print(act_type)
surfaces, features = analyze_morph(sent)
# print("----------surfaces,features--------------")
# print(features)
# print(surfaces)
morphed_sent = [[surfaces[i]] + features[i].split(',')
for i in range(len(surfaces))]
features = sent2features(morphed_sent)
print("----------morphed_sent,features--------------")
print(morphed_sent)
print(features)
named_entity = self.__extractor.extract(features, morphed_sent)
print("----------named_entity--------------")
print(named_entity)
dialogue_act = {'user_act_type': act_type}
if act_type != 'other':
dialogue_act.update(dict(named_entity))
#
# print("----------dialogue_act--------------")
# print(dialogue_act)
return dialogue_act
def execute(self, sent):
# 対話行為タイプの推定開始
# 4タイプ
# - genre: イタリアンとか
# - location: 新宿とか
# - money: 1万円とか
# - other: その他
features = sent2features_(sent)
#print("送信されたテキストから変換したfeatures:", features)
act_type = self.__predictor.predict([features])
print("featureから予測したact_type:", act_type)
# 対話行為タイプの推定完了
# 属性抽出
# ジャンルに対して文字を抽出。イタリアンとか中華とか
surfaces, features = analyze_morph(sent)
print("surfaces:", surfaces)
print("features:", features)
morphed_sent = [[surfaces[i]] + features[i].split(',')
for i in range(len(surfaces))]
print("morphed_sent", morphed_sent)
features = sent2features(morphed_sent)
named_entity = self.__extractor.extract(features, morphed_sent)
dialogue_act = {'user_act_type': act_type}
# ここで属性を追加
dialogue_act.update(dict(named_entity))
return dialogue_act
def sent2features_(sent):
from training_data_generator.scripts.analyzer import analyze_morph
dic_path = os.path.join(os.path.dirname(__file__), 'dic.txt')
surfaces, _ = analyze_morph(sent)
dictionary = corpora.Dictionary.load_from_text(dic_path)
features = to_features(dictionary, surfaces)
return features
def sent2features_(sent):
from training_data_generator.scripts.analyzer import analyze_morph
dic_path = os.path.join(os.path.dirname(__file__), 'dic.txt')
surfaces, _ = analyze_morph(sent)
dictionary = corpora.Dictionary.load_from_text(dic_path)
features = to_features(dictionary, surfaces)
return features
def sent2features_(sent):
from training_data_generator.scripts.analyzer import analyze_morph
dic_path = os.path.join(os.path.dirname(__file__), 'dic.txt')
# mecabで形態素解析して表層形を取得
surfaces, _ = analyze_morph(sent)
print("surfaces:", surfaces)
# コーパス
dictionary = corpora.Dictionary.load_from_text(dic_path)
print("dictionary: ", dictionary)
features = to_features(dictionary, surfaces)
return features
def execute(self, sent):
features = sent2features_(sent)
act_type = self.__predictor.predict([features])
surfaces, features = analyze_morph(sent)
morphed_sent = [[surfaces[i]] + features[i].split(',') for i in range(len(surfaces))]
features = sent2features(morphed_sent)
named_entity = self.__extractor.extract(features, morphed_sent)
dialogue_act = {'user_act_type': act_type}
dialogue_act.update(dict(named_entity))
return dialogue_act
def execute(self, sent):
features = sent2features_(sent)
act_type = self.__predictor.predict([features])
surfaces, features = analyze_morph(sent)
morphed_sent = [[surfaces[i]] + features[i].split(',')
for i in range(len(surfaces))]
features = sent2features(morphed_sent)
named_entity = self.__extractor.extract(features, morphed_sent)
dialogue_act = {'user_act_type': act_type}
dialogue_act.update(dict(named_entity))
return dialogue_act
def matching(sentence, ne_list):
def get_word_pos_list(sentence, ne_list):
word_pos_list = []
searched_pos = 0
for nw, ne in ne_list:
nw = ''.join(nw.split(' '))
idx = sentence.index(nw, searched_pos)
searched_pos = idx + len(nw)
word_pos_list.append((idx, searched_pos))
return word_pos_list
def get_morph_pos_list(wakati, word_pos_list):
morph_pos_list = []
for start_pos, end_pos in word_pos_list:
ch_cnt = 0
morph_pos = []
for i, morph in enumerate(wakati):
if ch_cnt < start_pos:
ch_cnt += len(morph)
elif ch_cnt < end_pos:
morph_pos.append(i)
ch_cnt += len(morph)
else:
break
morph_pos_list.append(morph_pos)
return morph_pos_list
def tagging(morph_pos_list, ne_list, ylabel):
for i in range(len(ne_list)):
nw, ne = ne_list[i]
morph_pos = morph_pos_list[i]
for j, k in enumerate(morph_pos):
prefix = 'B' if j == 0 else 'I'
ylabel[k] = prefix + '-' + ne
wakati, features = analyze_morph(sentence)
sentence = ''.join(sentence.split(' '))
word_pos_list = get_word_pos_list(sentence, ne_list)
morph_pos_list = get_morph_pos_list(wakati, word_pos_list)
ylabel = ['O'] * len(wakati)
tagging(morph_pos_list, ne_list, ylabel)
labeled_sent = []
for i in range(len(wakati)):
tmp = []
tmp.append(wakati[i])
tmp.extend(features[i].split(','))
tmp.append(ylabel[i])
labeled_sent.append(tmp)
# Error
for word_pos, morph_pos in zip(word_pos_list, morph_pos_list):
start_pos, end_pos = word_pos
word1, word2 = sentence[start_pos: end_pos], ''.join([wakati[i] for i in morph_pos])
if word1 != word2:
print(word1, word2)
return labeled_sent
def matching(sentence, ne_list):
def get_word_pos_list(sentence, ne_list):
word_pos_list = []
searched_pos = 0
for nw, ne in ne_list:
nw = ''.join(nw.split(' '))
idx = sentence.index(nw, searched_pos)
searched_pos = idx + len(nw)
word_pos_list.append((idx, searched_pos))
return word_pos_list
def get_morph_pos_list(wakati, word_pos_list):
morph_pos_list = []
for start_pos, end_pos in word_pos_list:
ch_cnt = 0
morph_pos = []
for i, morph in enumerate(wakati):
if ch_cnt < start_pos:
ch_cnt += len(morph)
elif ch_cnt < end_pos:
morph_pos.append(i)
ch_cnt += len(morph)
else:
break
morph_pos_list.append(morph_pos)
return morph_pos_list
def tagging(morph_pos_list, ne_list, ylabel):
for i in range(len(ne_list)):
nw, ne = ne_list[i]
morph_pos = morph_pos_list[i]
for j, k in enumerate(morph_pos):
prefix = 'B' if j == 0 else 'I'
ylabel[k] = prefix + '-' + ne
wakati, features = analyze_morph(sentence)
sentence = ''.join(sentence.split(' '))
word_pos_list = get_word_pos_list(sentence, ne_list)
morph_pos_list = get_morph_pos_list(wakati, word_pos_list)
ylabel = ['O'] * len(wakati)
tagging(morph_pos_list, ne_list, ylabel)
labeled_sent = []
for i in range(len(wakati)):
tmp = []
tmp.append(wakati[i])
tmp.extend(features[i].split(','))
tmp.append(ylabel[i])
labeled_sent.append(tmp)
# Error
for word_pos, morph_pos in zip(word_pos_list, morph_pos_list):
start_pos, end_pos = word_pos
word1, word2 = sentence[start_pos:end_pos], ''.join(
[wakati[i] for i in morph_pos])
if word1 != word2:
print(word1, word2)
return labeled_sent
