crosswoz / preprocess.py
zhuqi's picture
update crosswoz data
98b2db6
import copy
import json
import os
import re
from collections import Counter
from pprint import pprint
from shutil import copy2, rmtree
from zipfile import ZIP_DEFLATED, ZipFile
from tqdm import tqdm
ontology = {
"domains": {
"景点": {
"description": "查找景点",
"slots": {
"名称": {
"description": "景点名称",
"is_categorical": False,
"possible_values": []
},
"门票": {
"description": "景点门票价格",
"is_categorical": False,
"possible_values": []
},
"游玩时间": {
"description": "景点游玩时间",
"is_categorical": False,
"possible_values": []
},
"评分": {
"description": "景点评分",
"is_categorical": False,
"possible_values": []
},
"地址": {
"description": "景点地址",
"is_categorical": False,
"possible_values": []
},
"电话": {
"description": "景点电话",
"is_categorical": False,
"possible_values": []
},
"周边景点": {
"description": "景点周边景点",
"is_categorical": False,
"possible_values": []
},
"周边餐馆": {
"description": "景点周边餐馆",
"is_categorical": False,
"possible_values": []
},
"周边酒店": {
"description": "景点周边酒店",
"is_categorical": False,
"possible_values": []
}
}
},
"餐馆": {
"description": "查找餐馆",
"slots": {
"名称": {
"description": "餐馆名称",
"is_categorical": False,
"possible_values": []
},
"推荐菜": {
"description": "餐馆推荐菜",
"is_categorical": False,
"possible_values": []
},
"人均消费": {
"description": "餐馆人均消费",
"is_categorical": False,
"possible_values": []
},
"评分": {
"description": "餐馆评分",
"is_categorical": False,
"possible_values": []
},
"地址": {
"description": "餐馆地址",
"is_categorical": False,
"possible_values": []
},
"电话": {
"description": "餐馆电话",
"is_categorical": False,
"possible_values": []
},
"营业时间": {
"description": "餐馆营业时间",
"is_categorical": False,
"possible_values": []
},
"周边景点": {
"description": "餐馆周边景点",
"is_categorical": False,
"possible_values": []
},
"周边餐馆": {
"description": "餐馆周边餐馆",
"is_categorical": False,
"possible_values": []
},
"周边酒店": {
"description": "餐馆周边酒店",
"is_categorical": False,
"possible_values": []
}
}
},
"酒店": {
"description": "查找酒店",
"slots": {
"名称": {
"description": "酒店名称",
"is_categorical": False,
"possible_values": []
},
"酒店类型": {
"description": "酒店类型",
"is_categorical": True,
"possible_values": [
'高档型', '豪华型', '经济型', '舒适型'
]
},
"酒店设施": {
"description": "酒店设施",
"is_categorical": False,
"possible_values": []
},
"价格": {
"description": "酒店价格",
"is_categorical": False,
"possible_values": []
},
"评分": {
"description": "酒店评分",
"is_categorical": False,
"possible_values": []
},
"地址": {
"description": "酒店地址",
"is_categorical": False,
"possible_values": []
},
"电话": {
"description": "酒店电话",
"is_categorical": False,
"possible_values": []
},
"周边景点": {
"description": "酒店周边景点",
"is_categorical": False,
"possible_values": []
},
"周边餐馆": {
"description": "酒店周边餐馆",
"is_categorical": False,
"possible_values": []
},
"周边酒店": {
"description": "酒店周边酒店",
"is_categorical": False,
"possible_values": []
}
}
},
"地铁": {
"description": "乘坐地铁从某地到某地",
"slots": {
"出发地": {
"description": "地铁出发地",
"is_categorical": False,
"possible_values": []
},
"目的地": {
"description": "地铁目的地",
"is_categorical": False,
"possible_values": []
},
"出发地附近地铁站": {
"description": "出发地附近地铁站",
"is_categorical": False,
"possible_values": []
},
"目的地附近地铁站": {
"description": "目的地附近地铁站",
"is_categorical": False,
"possible_values": []
}
}
},
"出租": {
"description": "乘坐出租车从某地到某地",
"slots": {
"出发地": {
"description": "出租出发地",
"is_categorical": False,
"possible_values": []
},
"目的地": {
"description": "出租目的地",
"is_categorical": False,
"possible_values": []
},
"车型": {
"description": "出租车车型",
"is_categorical": True,
"possible_values": [
"#CX"
]
},
"车牌": {
"description": "出租车车牌",
"is_categorical": True,
"possible_values": [
"#CP"
]
}
}
},
"General": {
"description": "通用领域",
"slots": {}
}
},
"intents": {
"Inform": {
"description": "告知相关属性"
},
"Request": {
"description": "询问相关属性"
},
"Recommend": {
"description": "推荐搜索结果"
},
"Select": {
"description": "在附近搜索"
},
"NoOffer": {
"description": "未找到符合用户要求的结果"
},
"bye": {
"description": "再见"
},
"thank": {
"description": "感谢"
},
"welcome": {
"description": "不客气"
},
"greet": {
"description": "打招呼"
},
},
"state": {
"景点": {
"名称": "",
"门票": "",
"游玩时间": "",
"评分": "",
"周边景点": "",
"周边餐馆": "",
"周边酒店": "",
},
"餐馆": {
"名称": "",
"推荐菜": "",
"人均消费": "",
"评分": "",
"周边景点": "",
"周边餐馆": "",
"周边酒店": "",
},
"酒店": {
"名称": "",
"酒店类型": "",
"酒店设施": "",
"价格": "",
"评分": "",
"周边景点": "",
"周边餐馆": "",
"周边酒店": "",
},
"地铁": {
"出发地": "",
"目的地": "",
},
"出租": {
"出发地": "",
"目的地": "",
}
},
"dialogue_acts": {
"categorical": {},
"non-categorical": {},
"binary": {}
}
}
cnt_domain_slot = Counter()
def convert_da(da_list, utt):
'''
convert dialogue acts to required format
:param da_dict: list of (intent, domain, slot, value)
:param utt: user or system utt
'''
global ontology, cnt_domain_slot
converted_da = {
'categorical': [],
'non-categorical': [],
'binary': []
}
for intent, domain, slot, value in da_list:
# if intent in ['Inform', 'Recommend']:
if intent == 'NoOffer':
assert slot == 'none' and value == 'none'
converted_da['binary'].append({
'intent': intent,
'domain': domain,
'slot': ''
})
elif intent == 'General':
# intent=General, domain=thank/bye/greet/welcome
assert slot == 'none' and value == 'none'
converted_da['binary'].append({
'intent': domain,
'domain': intent,
'slot': ''
})
elif intent == 'Request':
assert value == '' and slot != 'none'
converted_da['binary'].append({
'intent': intent,
'domain': domain,
'slot': slot
})
elif '酒店设施' in slot:
converted_da['binary'].append({
'intent': intent,
'domain': domain,
'slot': f"{slot}-{value}"
})
elif intent == 'Select':
assert slot == '源领域'
converted_da['binary'].append({
'intent': intent,
'domain': domain,
'slot': f"{slot}-{value}"
})
elif slot in ['酒店类型', '车型', '车牌']:
assert intent in ['Inform', 'Recommend']
assert slot != 'none' and value != 'none'
converted_da['categorical'].append({
'intent': intent,
'domain': domain,
'slot': slot,
'value': value
})
else:
assert intent in ['Inform', 'Recommend']
assert slot != 'none' and value != 'none'
matches = utt.count(value)
if matches == 1:
start = utt.index(value)
end = start + len(value)
converted_da['non-categorical'].append({
'intent': intent,
'domain': domain,
'slot': slot,
'value': value,
'start': start,
'end': end
})
cnt_domain_slot['have span'] += 1
else:
# can not find span
converted_da['non-categorical'].append({
'intent': intent,
'domain': domain,
'slot': slot,
'value': value
})
cnt_domain_slot['no span'] += 1
# cnt_domain_slot.setdefault(f'{domain}-{slot}', set())
# cnt_domain_slot[f'{domain}-{slot}'].add(value)
return converted_da
def transform_user_state(user_state):
goal = []
for subgoal in user_state:
gid, domain, slot, value, mentioned = subgoal
if len(value) != 0:
t = 'inform'
else:
t = 'request'
if len(goal) < gid:
goal.append({domain: {'inform': {}, 'request': {}}})
goal[gid-1][domain][t][slot] = [value, 'mentioned' if mentioned else 'not mentioned']
return goal
def preprocess():
original_data_dir = '../../crosswoz'
new_data_dir = 'data'
os.makedirs(new_data_dir, exist_ok=True)
for filename in os.listdir(os.path.join(original_data_dir,'database')):
copy2(f'{original_data_dir}/database/{filename}', new_data_dir)
global ontology
dataset = 'crosswoz'
splits = ['train', 'validation', 'test']
dialogues_by_split = {split: [] for split in splits}
for split in ['train', 'val', 'test']:
data = json.load(ZipFile(os.path.join(original_data_dir, f'{split}.json.zip'), 'r').open(f'{split}.json'))
if split == 'val':
split = 'validation'
for ori_dialog_id, ori_dialog in data.items():
dialogue_id = f'{dataset}-{split}-{len(dialogues_by_split[split])}'
# get user goal and involved domains
goal = {'inform': {}, 'request': {}}
goal["description"] = '\n'.join(ori_dialog["task description"])
cur_domains = [x[1] for i, x in enumerate(ori_dialog['goal']) if i == 0 or ori_dialog['goal'][i-1][1] != x[1]]
dialogue = {
'dataset': dataset,
'data_split': split,
'dialogue_id': dialogue_id,
'original_id': ori_dialog_id,
'domains': cur_domains,
'goal': goal,
'user_state_init': transform_user_state(ori_dialog['goal']),
'type': ori_dialog['type'],
'turns': [],
'user_state_final': transform_user_state(ori_dialog['final_goal'])
}
for turn_id, turn in enumerate(ori_dialog['messages']):
# skip error turns
if ori_dialog_id == '2660' and turn_id in [8,9]:
continue
elif ori_dialog_id == '7467' and turn_id in [14,15]:
continue
elif ori_dialog_id == '11726' and turn_id in [4,5]:
continue
elif ori_dialog_id == '10550' and turn_id == 6:
dialogue['user_state_final'] = dialogue['turns'][-2]['user_state']
break
elif ori_dialog_id == '11724' and turn_id == 8:
dialogue['user_state_final'] = dialogue['turns'][-2]['user_state']
break
speaker = 'user' if turn['role'] == 'usr' else 'system'
utt = turn['content']
das = turn['dialog_act']
dialogue_acts = convert_da(das, utt)
dialogue['turns'].append({
'speaker': speaker,
'utterance': utt,
'utt_idx': len(dialogue['turns']),
'dialogue_acts': dialogue_acts,
})
# add to dialogue_acts dictionary in the ontology
for da_type in dialogue_acts:
das = dialogue_acts[da_type]
for da in das:
ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])][speaker] = True
if speaker == 'user':
dialogue['turns'][-1]['user_state'] = transform_user_state(turn['user_state'])
else:
# add state to last user turn
belief_state = turn['sys_state_init']
for domain in belief_state:
belief_state[domain].pop('selectedResults')
dialogue['turns'][-2]['state'] = belief_state
db_query = turn['sys_state']
db_results = {}
for domain in list(db_query.keys()):
db_res = db_query[domain].pop('selectedResults')
if len(db_res) > 0:
db_results[domain] = [{'名称': x} for x in db_res]
else:
db_query.pop(domain)
dialogue['turns'][-1]['db_query'] = db_query
dialogue['turns'][-1]['db_results'] = db_results
dialogues_by_split[split].append(dialogue)
pprint(cnt_domain_slot.most_common())
dialogues = []
for split in splits:
dialogues += dialogues_by_split[split]
for da_type in ontology['dialogue_acts']:
ontology["dialogue_acts"][da_type] = sorted([str(
{'user': speakers.get('user', False), 'system': speakers.get('system', False), 'intent': da[0],
'domain': da[1], 'slot': da[2]}) for da, speakers in ontology["dialogue_acts"][da_type].items()])
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
for filename in os.listdir(new_data_dir):
zf.write(f'{new_data_dir}/{filename}')
rmtree(new_data_dir)
return dialogues, ontology
def fix_entity_booked_info(entity_booked_dict, booked):
for domain in entity_booked_dict:
if not entity_booked_dict[domain] and booked[domain]:
entity_booked_dict[domain] = True
booked[domain] = []
return entity_booked_dict, booked
if __name__ == '__main__':
preprocess()