Upload 6 files
Browse files- README.md +61 -3
- data.zip +3 -0
- database.py +239 -0
- dummy_data.json +0 -0
- preprocess.py +535 -0
- shuffled_dial_ids.json +0 -0
README.md
CHANGED
@@ -1,3 +1,61 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Dataset Card for CrossWOZ
|
2 |
+
|
3 |
+
- **Repository:** https://github.com/thu-coai/CrossWOZ
|
4 |
+
- **Paper:** https://aclanthology.org/2020.tacl-1.19/
|
5 |
+
- **Leaderboard:** None
|
6 |
+
- **Who transforms the dataset:** Qi Zhu(zhuq96 at gmail dot com)
|
7 |
+
|
8 |
+
### Dataset Summary
|
9 |
+
|
10 |
+
CrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides. We also provide a user simulator and several benchmark models for pipelined taskoriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus.
|
11 |
+
|
12 |
+
- **How to get the transformed data from original data:**
|
13 |
+
- Run `python preprocess.py` in the current directory. Need `../../crosswoz/` as the original data.
|
14 |
+
- **Main changes of the transformation:**
|
15 |
+
- Add simple description for domains, slots, and intents.
|
16 |
+
- switch intent&domain of `General` dialog acts => domain == 'General' and intent in ['thank','bye','greet','welcome']
|
17 |
+
- Binary dialog acts include: 1) domain == 'General'; 2) intent in ['NoOffer', 'Request', 'Select']; 3) slot in ['酒店设施']
|
18 |
+
- Categorical dialog acts include: slot in ['酒店类型', '车型', '车牌']
|
19 |
+
- Non-categorical dialogue acts: others. assert intent in ['Inform', 'Recommend'] and slot != 'none' and value != 'none'
|
20 |
+
- Transform original user goal to list of `{domain: {'inform': {slot: [value, mentioned/not mentioned]}, 'request': {slot: [value, mentioned/not mentioned]}}}`, stored as `user_state` of user turns.
|
21 |
+
- Transform `sys_state_init` (first API call of system turns) without `selectedResults` as belief state in user turns.
|
22 |
+
- Transform `sys_state` (last API call of system turns) to `db_query` with domain states that contain non-empty `selectedResults`. The `selectedResults` are saved as `db_results`. Both stored in system turns.
|
23 |
+
- **Annotations:**
|
24 |
+
- user goal, user state, dialogue acts, state, db query, db results.
|
25 |
+
- Multiple values in state are separated by spaces, meaning all constraints should be satisfied.
|
26 |
+
|
27 |
+
### Supported Tasks and Leaderboards
|
28 |
+
|
29 |
+
NLU, DST, Policy, NLG, E2E, User simulator
|
30 |
+
|
31 |
+
### Languages
|
32 |
+
|
33 |
+
Chinese
|
34 |
+
|
35 |
+
### Data Splits
|
36 |
+
|
37 |
+
| split | dialogues | utterances | avg_utt | avg_tokens | avg_domains | cat slot match(state) | cat slot match(goal) | cat slot match(dialogue act) | non-cat slot span(dialogue act) |
|
38 |
+
|------------|-------------|--------------|-----------|--------------|---------------|-------------------------|------------------------|--------------------------------|-----------------------------------|
|
39 |
+
| train | 5010 | 84660 | 16.9 | 20.55 | 3.02 | 99.67 | - | 100 | 94.39 |
|
40 |
+
| validation | 500 | 8458 | 16.92 | 20.53 | 3.04 | 99.62 | - | 100 | 94.36 |
|
41 |
+
| test | 500 | 8476 | 16.95 | 20.51 | 3.08 | 99.61 | - | 100 | 94.85 |
|
42 |
+
| all | 6010 | 101594 | 16.9 | 20.54 | 3.03 | 99.66 | - | 100 | 94.43 |
|
43 |
+
|
44 |
+
6 domains: ['景点', '餐馆', '酒店', '地铁', '出租', 'General']
|
45 |
+
- **cat slot match**: how many values of categorical slots are in the possible values of ontology in percentage.
|
46 |
+
- **non-cat slot span**: how many values of non-categorical slots have span annotation in percentage.
|
47 |
+
|
48 |
+
### Citation
|
49 |
+
|
50 |
+
```
|
51 |
+
@article{zhu2020crosswoz,
|
52 |
+
author = {Qi Zhu and Kaili Huang and Zheng Zhang and Xiaoyan Zhu and Minlie Huang},
|
53 |
+
title = {Cross{WOZ}: A Large-Scale Chinese Cross-Domain Task-Oriented Dialogue Dataset},
|
54 |
+
journal = {Transactions of the Association for Computational Linguistics},
|
55 |
+
year = {2020}
|
56 |
+
}
|
57 |
+
```
|
58 |
+
|
59 |
+
### Licensing Information
|
60 |
+
|
61 |
+
Apache License, Version 2.0
|
data.zip
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:ddc310bba9fada91f266bd455d50b064ec897f069b960c6ca63447cefb8355bb
|
3 |
+
size 15918515
|
database.py
ADDED
@@ -0,0 +1,239 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""Modifed from https://github.com/thu-coai/CrossWOZ/blob/master/convlab2/util/crosswoz/dbquery.py"""
|
2 |
+
import json
|
3 |
+
import os
|
4 |
+
import re
|
5 |
+
from zipfile import ZipFile
|
6 |
+
|
7 |
+
from convlab.util.unified_datasets_util import (BaseDatabase,
|
8 |
+
download_unified_datasets)
|
9 |
+
|
10 |
+
|
11 |
+
def contains(arr, s):
|
12 |
+
return not len(tuple(filter(lambda item: (not (item.find(s) < 0)), arr)))
|
13 |
+
|
14 |
+
class Database(BaseDatabase):
|
15 |
+
def __init__(self):
|
16 |
+
"""extract data.zip and load the database."""
|
17 |
+
data_path = download_unified_datasets('crosswoz', 'data.zip', os.path.dirname(os.path.abspath(__file__)))
|
18 |
+
archive = ZipFile(data_path)
|
19 |
+
self.domains = ['景点', '餐馆', '酒店', '地铁', '出租']
|
20 |
+
domain2eng = {'景点': 'attraction', '餐馆': 'restaurant', '酒店': 'hotel', '地铁': 'metro'}
|
21 |
+
self.data = {}
|
22 |
+
for domain in self.domains[:-1]:
|
23 |
+
with archive.open('data/{}_db.json'.format(domain2eng[domain])) as f:
|
24 |
+
self.data[domain] = json.loads(f.read())
|
25 |
+
|
26 |
+
self.schema = {
|
27 |
+
'景点': {
|
28 |
+
'名称': {'params': None},
|
29 |
+
'门票': {'type': 'between', 'params': [None, None]},
|
30 |
+
'游玩时间': {'params': None},
|
31 |
+
'评分': {'type': 'between', 'params': [None, None]},
|
32 |
+
'周边景点': {'type': 'in', 'params': None},
|
33 |
+
'周边餐馆': {'type': 'in', 'params': None},
|
34 |
+
'周边酒店': {'type': 'in', 'params': None},
|
35 |
+
},
|
36 |
+
'餐馆': {
|
37 |
+
'名称': {'params': None},
|
38 |
+
'推荐菜': {'type': 'multiple_in', 'params': None},
|
39 |
+
'人均消费': {'type': 'between', 'params': [None, None]},
|
40 |
+
'评分': {'type': 'between', 'params': [None, None]},
|
41 |
+
'周边景点': {'type': 'in', 'params': None},
|
42 |
+
'周边餐馆': {'type': 'in', 'params': None},
|
43 |
+
'周边酒店': {'type': 'in', 'params': None}
|
44 |
+
},
|
45 |
+
'酒店': {
|
46 |
+
'名称': {'params': None},
|
47 |
+
'酒店类型': {'params': None},
|
48 |
+
'酒店设施': {'type': 'multiple_in', 'params': None},
|
49 |
+
'价格': {'type': 'between', 'params': [None, None]},
|
50 |
+
'评分': {'type': 'between', 'params': [None, None]},
|
51 |
+
'周边景点': {'type': 'in', 'params': None},
|
52 |
+
'周边餐馆': {'type': 'in', 'params': None},
|
53 |
+
'周边酒店': {'type': 'in', 'params': None}
|
54 |
+
},
|
55 |
+
'地铁': {
|
56 |
+
'起点': {'params': None},
|
57 |
+
'终点': {'params': None},
|
58 |
+
},
|
59 |
+
'出租': {
|
60 |
+
'起点': {'params': None},
|
61 |
+
'终点': {'params': None},
|
62 |
+
}
|
63 |
+
}
|
64 |
+
|
65 |
+
def query(self, domain: str, state: dict, topk: int) -> list:
|
66 |
+
"""
|
67 |
+
return a list of topk entities (dict containing slot-value pairs) for a given domain based on the dialogue state.
|
68 |
+
query database using belief state, return list of entities, same format as database
|
69 |
+
:param state: belief state of the format {domain: {slot: value}}
|
70 |
+
:param domain: maintain by DST, current query domain
|
71 |
+
:param topk: max number of entities
|
72 |
+
:return: list of entities
|
73 |
+
"""
|
74 |
+
if not domain:
|
75 |
+
return []
|
76 |
+
cur_query_form = {}
|
77 |
+
for slot, value in state[domain].items():
|
78 |
+
if not value:
|
79 |
+
continue
|
80 |
+
if slot == '出发地':
|
81 |
+
slot = '起点'
|
82 |
+
elif slot == '目的地':
|
83 |
+
slot = '终点'
|
84 |
+
if slot == '名称':
|
85 |
+
# DONE: if name is specified, ignore other constraints
|
86 |
+
cur_query_form = {'名称': value}
|
87 |
+
break
|
88 |
+
elif slot == '评分':
|
89 |
+
if re.match('(\d\.\d|\d)', value):
|
90 |
+
if re.match('\d\.\d', value):
|
91 |
+
score = float(re.match('\d\.\d', value)[0])
|
92 |
+
else:
|
93 |
+
score = int(re.match('\d', value)[0])
|
94 |
+
cur_query_form[slot] = [score, None]
|
95 |
+
# else:
|
96 |
+
# assert 0, value
|
97 |
+
elif slot in ['门票', '人均消费', '价格']:
|
98 |
+
low, high = None, None
|
99 |
+
if re.match('(\d+)-(\d+)', value):
|
100 |
+
low = int(re.match('(\d+)-(\d+)', value)[1])
|
101 |
+
high = int(re.match('(\d+)-(\d+)', value)[2])
|
102 |
+
elif re.match('\d+', value):
|
103 |
+
if '以上' in value:
|
104 |
+
low = int(re.match('\d+', value)[0])
|
105 |
+
elif '以下' in value:
|
106 |
+
high = int(re.match('\d+', value)[0])
|
107 |
+
else:
|
108 |
+
low = high = int(re.match('\d+', value)[0])
|
109 |
+
elif slot == '门票':
|
110 |
+
if value == '免费':
|
111 |
+
low = high = 0
|
112 |
+
elif value == '不免费':
|
113 |
+
low = 1
|
114 |
+
else:
|
115 |
+
print(value)
|
116 |
+
# assert 0
|
117 |
+
cur_query_form[slot] = [low, high]
|
118 |
+
else:
|
119 |
+
cur_query_form[slot] = value
|
120 |
+
cur_res = self.query_schema(field=domain, args=cur_query_form)
|
121 |
+
if domain == '出租':
|
122 |
+
res = [cur_res]
|
123 |
+
elif domain == '地铁':
|
124 |
+
res = []
|
125 |
+
for r in cur_res:
|
126 |
+
if not res and '起点' in r[0]:
|
127 |
+
res.append(r)
|
128 |
+
break
|
129 |
+
for r in cur_res:
|
130 |
+
if '终点' in r[0]:
|
131 |
+
res.append(r)
|
132 |
+
break
|
133 |
+
else:
|
134 |
+
res = cur_res
|
135 |
+
|
136 |
+
return res[:topk]
|
137 |
+
|
138 |
+
def query_schema(self, field, args):
|
139 |
+
if not field in self.schema:
|
140 |
+
raise Exception('Unknown field %s' % field)
|
141 |
+
if not isinstance(args, dict):
|
142 |
+
raise Exception('`args` must be dict')
|
143 |
+
db = self.data.get(field)
|
144 |
+
plan = self.schema[field]
|
145 |
+
for key, value in args.items():
|
146 |
+
if not key in plan:
|
147 |
+
raise Exception('Unknown key %s' % key)
|
148 |
+
value_type = plan[key].get('type')
|
149 |
+
if value_type == 'between':
|
150 |
+
if not value[0] is None:
|
151 |
+
plan[key]['params'][0] = float(value[0])
|
152 |
+
if not value[1] is None:
|
153 |
+
plan[key]['params'][1] = float(value[1])
|
154 |
+
else:
|
155 |
+
if not isinstance(value, str):
|
156 |
+
raise Exception('Value for `%s` must be string' % key)
|
157 |
+
plan[key]['params'] = value
|
158 |
+
if field in ['地铁', '出租']:
|
159 |
+
s = plan['起点']['params']
|
160 |
+
e = plan['终点']['params']
|
161 |
+
if not s or not e:
|
162 |
+
return []
|
163 |
+
if field == '出租':
|
164 |
+
return [
|
165 |
+
'出租 (%s - %s)' % (s, e), {
|
166 |
+
'领域': '出租',
|
167 |
+
'车型': '#CX',
|
168 |
+
'车牌': '#CP'
|
169 |
+
}
|
170 |
+
]
|
171 |
+
else:
|
172 |
+
def func1(item):
|
173 |
+
if item[0].find(s) >= 0:
|
174 |
+
return ['(起点) %s' % item[0], item[1]]
|
175 |
+
|
176 |
+
def func2(item):
|
177 |
+
if item[0].find(e) >= 0:
|
178 |
+
return ['(终点) %s' % item[0], item[1]]
|
179 |
+
return None
|
180 |
+
|
181 |
+
return list(filter(lambda item: not item is None, list(map(func1, db)))) + list(
|
182 |
+
filter(lambda item: not item is None, list(map(func2, db))))
|
183 |
+
|
184 |
+
def func3(item):
|
185 |
+
details = item[1]
|
186 |
+
for key, val in args.items():
|
187 |
+
val = details.get(key)
|
188 |
+
absence = val is None
|
189 |
+
options = plan[key]
|
190 |
+
if options.get('type') == 'between':
|
191 |
+
L = options['params'][0]
|
192 |
+
R = options['params'][1]
|
193 |
+
if not L is None:
|
194 |
+
if absence:
|
195 |
+
return False
|
196 |
+
else:
|
197 |
+
L = float('-inf')
|
198 |
+
if not R is None:
|
199 |
+
if absence:
|
200 |
+
return False
|
201 |
+
else:
|
202 |
+
R = float('inf')
|
203 |
+
if L > val or val > R:
|
204 |
+
return False
|
205 |
+
elif options.get('type') == 'in':
|
206 |
+
s = options['params']
|
207 |
+
if not s is None:
|
208 |
+
if absence:
|
209 |
+
return False
|
210 |
+
if contains(val, s):
|
211 |
+
return False
|
212 |
+
elif options.get('type') == 'multiple_in':
|
213 |
+
s = options['params']
|
214 |
+
if not s is None:
|
215 |
+
if absence:
|
216 |
+
return False
|
217 |
+
sarr = list(filter(lambda t: bool(t), s.split(' ')))
|
218 |
+
if len(list(filter(lambda t: contains(val, t), sarr))):
|
219 |
+
return False
|
220 |
+
else:
|
221 |
+
s = options['params']
|
222 |
+
if not s is None:
|
223 |
+
if absence:
|
224 |
+
return False
|
225 |
+
if val.find(s) < 0:
|
226 |
+
return False
|
227 |
+
return True
|
228 |
+
|
229 |
+
return [x[1] for x in filter(func3, db)]
|
230 |
+
|
231 |
+
|
232 |
+
if __name__ == '__main__':
|
233 |
+
db = Database()
|
234 |
+
assert issubclass(Database, BaseDatabase)
|
235 |
+
assert isinstance(db, BaseDatabase)
|
236 |
+
res = db.query("餐馆", {"餐馆":{'评分':'4.5以上'}}, topk=3)
|
237 |
+
from pprint import pprint
|
238 |
+
pprint(res)
|
239 |
+
print(len(res))
|
dummy_data.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
preprocess.py
ADDED
@@ -0,0 +1,535 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import copy
|
2 |
+
import json
|
3 |
+
import os
|
4 |
+
import re
|
5 |
+
from collections import Counter
|
6 |
+
from pprint import pprint
|
7 |
+
from shutil import copy2, rmtree
|
8 |
+
from zipfile import ZIP_DEFLATED, ZipFile
|
9 |
+
|
10 |
+
from tqdm import tqdm
|
11 |
+
|
12 |
+
ontology = {
|
13 |
+
"domains": {
|
14 |
+
"景点": {
|
15 |
+
"description": "查找景点",
|
16 |
+
"slots": {
|
17 |
+
"名称": {
|
18 |
+
"description": "景点名称",
|
19 |
+
"is_categorical": False,
|
20 |
+
"possible_values": []
|
21 |
+
},
|
22 |
+
"门票": {
|
23 |
+
"description": "景点门票价格",
|
24 |
+
"is_categorical": False,
|
25 |
+
"possible_values": []
|
26 |
+
},
|
27 |
+
"游玩时间": {
|
28 |
+
"description": "景点游玩时间",
|
29 |
+
"is_categorical": False,
|
30 |
+
"possible_values": []
|
31 |
+
},
|
32 |
+
"评分": {
|
33 |
+
"description": "景点评分",
|
34 |
+
"is_categorical": False,
|
35 |
+
"possible_values": []
|
36 |
+
},
|
37 |
+
"地址": {
|
38 |
+
"description": "景点地址",
|
39 |
+
"is_categorical": False,
|
40 |
+
"possible_values": []
|
41 |
+
},
|
42 |
+
"电话": {
|
43 |
+
"description": "景点电话",
|
44 |
+
"is_categorical": False,
|
45 |
+
"possible_values": []
|
46 |
+
},
|
47 |
+
"周边景点": {
|
48 |
+
"description": "景点周边景点",
|
49 |
+
"is_categorical": False,
|
50 |
+
"possible_values": []
|
51 |
+
},
|
52 |
+
"周边餐馆": {
|
53 |
+
"description": "景点周边餐馆",
|
54 |
+
"is_categorical": False,
|
55 |
+
"possible_values": []
|
56 |
+
},
|
57 |
+
"周边酒店": {
|
58 |
+
"description": "景点周边酒店",
|
59 |
+
"is_categorical": False,
|
60 |
+
"possible_values": []
|
61 |
+
}
|
62 |
+
}
|
63 |
+
},
|
64 |
+
"餐馆": {
|
65 |
+
"description": "查找餐馆",
|
66 |
+
"slots": {
|
67 |
+
"名称": {
|
68 |
+
"description": "餐馆名称",
|
69 |
+
"is_categorical": False,
|
70 |
+
"possible_values": []
|
71 |
+
},
|
72 |
+
"推荐菜": {
|
73 |
+
"description": "餐馆推荐菜",
|
74 |
+
"is_categorical": False,
|
75 |
+
"possible_values": []
|
76 |
+
},
|
77 |
+
"人均消费": {
|
78 |
+
"description": "餐馆人均消费",
|
79 |
+
"is_categorical": False,
|
80 |
+
"possible_values": []
|
81 |
+
},
|
82 |
+
"评分": {
|
83 |
+
"description": "餐馆评分",
|
84 |
+
"is_categorical": False,
|
85 |
+
"possible_values": []
|
86 |
+
},
|
87 |
+
"地址": {
|
88 |
+
"description": "餐馆地址",
|
89 |
+
"is_categorical": False,
|
90 |
+
"possible_values": []
|
91 |
+
},
|
92 |
+
"电话": {
|
93 |
+
"description": "餐馆电话",
|
94 |
+
"is_categorical": False,
|
95 |
+
"possible_values": []
|
96 |
+
},
|
97 |
+
"营业时间": {
|
98 |
+
"description": "餐馆营业时间",
|
99 |
+
"is_categorical": False,
|
100 |
+
"possible_values": []
|
101 |
+
},
|
102 |
+
"周边景点": {
|
103 |
+
"description": "餐馆周边景点",
|
104 |
+
"is_categorical": False,
|
105 |
+
"possible_values": []
|
106 |
+
},
|
107 |
+
"周边餐馆": {
|
108 |
+
"description": "餐馆周边餐馆",
|
109 |
+
"is_categorical": False,
|
110 |
+
"possible_values": []
|
111 |
+
},
|
112 |
+
"周边酒店": {
|
113 |
+
"description": "餐馆周边酒店",
|
114 |
+
"is_categorical": False,
|
115 |
+
"possible_values": []
|
116 |
+
}
|
117 |
+
}
|
118 |
+
},
|
119 |
+
"酒店": {
|
120 |
+
"description": "查找酒店",
|
121 |
+
"slots": {
|
122 |
+
"名称": {
|
123 |
+
"description": "酒店名称",
|
124 |
+
"is_categorical": False,
|
125 |
+
"possible_values": []
|
126 |
+
},
|
127 |
+
"酒店类型": {
|
128 |
+
"description": "酒店类型",
|
129 |
+
"is_categorical": True,
|
130 |
+
"possible_values": [
|
131 |
+
'高档型', '豪华型', '经济型', '舒适型'
|
132 |
+
]
|
133 |
+
},
|
134 |
+
"酒店设施": {
|
135 |
+
"description": "酒店设施",
|
136 |
+
"is_categorical": False,
|
137 |
+
"possible_values": []
|
138 |
+
},
|
139 |
+
"价格": {
|
140 |
+
"description": "酒店价格",
|
141 |
+
"is_categorical": False,
|
142 |
+
"possible_values": []
|
143 |
+
},
|
144 |
+
"评分": {
|
145 |
+
"description": "酒店评分",
|
146 |
+
"is_categorical": False,
|
147 |
+
"possible_values": []
|
148 |
+
},
|
149 |
+
"地址": {
|
150 |
+
"description": "酒店地址",
|
151 |
+
"is_categorical": False,
|
152 |
+
"possible_values": []
|
153 |
+
},
|
154 |
+
"电话": {
|
155 |
+
"description": "酒店电话",
|
156 |
+
"is_categorical": False,
|
157 |
+
"possible_values": []
|
158 |
+
},
|
159 |
+
"周边景点": {
|
160 |
+
"description": "酒店周边景点",
|
161 |
+
"is_categorical": False,
|
162 |
+
"possible_values": []
|
163 |
+
},
|
164 |
+
"周边餐馆": {
|
165 |
+
"description": "酒店周边餐馆",
|
166 |
+
"is_categorical": False,
|
167 |
+
"possible_values": []
|
168 |
+
},
|
169 |
+
"周边酒店": {
|
170 |
+
"description": "酒店周边酒店",
|
171 |
+
"is_categorical": False,
|
172 |
+
"possible_values": []
|
173 |
+
}
|
174 |
+
}
|
175 |
+
},
|
176 |
+
"地铁": {
|
177 |
+
"description": "乘坐地铁从某地到某地",
|
178 |
+
"slots": {
|
179 |
+
"出发地": {
|
180 |
+
"description": "地铁出发地",
|
181 |
+
"is_categorical": False,
|
182 |
+
"possible_values": []
|
183 |
+
},
|
184 |
+
"目的地": {
|
185 |
+
"description": "地铁目的地",
|
186 |
+
"is_categorical": False,
|
187 |
+
"possible_values": []
|
188 |
+
},
|
189 |
+
"出发地附近地铁站": {
|
190 |
+
"description": "出发地附近地铁站",
|
191 |
+
"is_categorical": False,
|
192 |
+
"possible_values": []
|
193 |
+
},
|
194 |
+
"目的地附近地铁站": {
|
195 |
+
"description": "目的地附近地铁站",
|
196 |
+
"is_categorical": False,
|
197 |
+
"possible_values": []
|
198 |
+
}
|
199 |
+
}
|
200 |
+
},
|
201 |
+
"出租": {
|
202 |
+
"description": "乘坐出租车从某地到某地",
|
203 |
+
"slots": {
|
204 |
+
"出发地": {
|
205 |
+
"description": "出租出发地",
|
206 |
+
"is_categorical": False,
|
207 |
+
"possible_values": []
|
208 |
+
},
|
209 |
+
"目的地": {
|
210 |
+
"description": "出租目的地",
|
211 |
+
"is_categorical": False,
|
212 |
+
"possible_values": []
|
213 |
+
},
|
214 |
+
"车型": {
|
215 |
+
"description": "出租车车型",
|
216 |
+
"is_categorical": True,
|
217 |
+
"possible_values": [
|
218 |
+
"#CX"
|
219 |
+
]
|
220 |
+
},
|
221 |
+
"车牌": {
|
222 |
+
"description": "出租车车牌",
|
223 |
+
"is_categorical": True,
|
224 |
+
"possible_values": [
|
225 |
+
"#CP"
|
226 |
+
]
|
227 |
+
}
|
228 |
+
}
|
229 |
+
},
|
230 |
+
"General": {
|
231 |
+
"description": "通用领域",
|
232 |
+
"slots": {}
|
233 |
+
}
|
234 |
+
},
|
235 |
+
"intents": {
|
236 |
+
"Inform": {
|
237 |
+
"description": "告知相关属性"
|
238 |
+
},
|
239 |
+
"Request": {
|
240 |
+
"description": "询问相关属性"
|
241 |
+
},
|
242 |
+
"Recommend": {
|
243 |
+
"description": "推荐搜索结果"
|
244 |
+
},
|
245 |
+
"Select": {
|
246 |
+
"description": "在附近搜索"
|
247 |
+
},
|
248 |
+
"NoOffer": {
|
249 |
+
"description": "未找到符合用户要求的结果"
|
250 |
+
},
|
251 |
+
"bye": {
|
252 |
+
"description": "再见"
|
253 |
+
},
|
254 |
+
"thank": {
|
255 |
+
"description": "感谢"
|
256 |
+
},
|
257 |
+
"welcome": {
|
258 |
+
"description": "不客气"
|
259 |
+
},
|
260 |
+
"greet": {
|
261 |
+
"description": "打招呼"
|
262 |
+
},
|
263 |
+
},
|
264 |
+
"state": {
|
265 |
+
"景点": {
|
266 |
+
"名称": "",
|
267 |
+
"门票": "",
|
268 |
+
"游玩时间": "",
|
269 |
+
"评分": "",
|
270 |
+
"周边景点": "",
|
271 |
+
"周边餐馆": "",
|
272 |
+
"周边酒店": "",
|
273 |
+
},
|
274 |
+
"餐馆": {
|
275 |
+
"名称": "",
|
276 |
+
"推荐菜": "",
|
277 |
+
"人均消费": "",
|
278 |
+
"评分": "",
|
279 |
+
"周边景点": "",
|
280 |
+
"周边餐馆": "",
|
281 |
+
"周边酒店": "",
|
282 |
+
},
|
283 |
+
"酒店": {
|
284 |
+
"名称": "",
|
285 |
+
"酒店类型": "",
|
286 |
+
"酒店设施": "",
|
287 |
+
"价格": "",
|
288 |
+
"评分": "",
|
289 |
+
"周边景点": "",
|
290 |
+
"周边餐馆": "",
|
291 |
+
"周边酒店": "",
|
292 |
+
},
|
293 |
+
"地铁": {
|
294 |
+
"出发地": "",
|
295 |
+
"目的地": "",
|
296 |
+
},
|
297 |
+
"出租": {
|
298 |
+
"出发地": "",
|
299 |
+
"目的地": "",
|
300 |
+
}
|
301 |
+
},
|
302 |
+
"dialogue_acts": {
|
303 |
+
"categorical": {},
|
304 |
+
"non-categorical": {},
|
305 |
+
"binary": {}
|
306 |
+
}
|
307 |
+
}
|
308 |
+
|
309 |
+
cnt_domain_slot = Counter()
|
310 |
+
|
311 |
+
def convert_da(da_list, utt):
|
312 |
+
'''
|
313 |
+
convert dialogue acts to required format
|
314 |
+
:param da_dict: list of (intent, domain, slot, value)
|
315 |
+
:param utt: user or system utt
|
316 |
+
'''
|
317 |
+
global ontology, cnt_domain_slot
|
318 |
+
|
319 |
+
converted_da = {
|
320 |
+
'categorical': [],
|
321 |
+
'non-categorical': [],
|
322 |
+
'binary': []
|
323 |
+
}
|
324 |
+
|
325 |
+
for intent, domain, slot, value in da_list:
|
326 |
+
# if intent in ['Inform', 'Recommend']:
|
327 |
+
if intent == 'NoOffer':
|
328 |
+
assert slot == 'none' and value == 'none'
|
329 |
+
converted_da['binary'].append({
|
330 |
+
'intent': intent,
|
331 |
+
'domain': domain,
|
332 |
+
'slot': ''
|
333 |
+
})
|
334 |
+
elif intent == 'General':
|
335 |
+
# intent=General, domain=thank/bye/greet/welcome
|
336 |
+
assert slot == 'none' and value == 'none'
|
337 |
+
converted_da['binary'].append({
|
338 |
+
'intent': domain,
|
339 |
+
'domain': intent,
|
340 |
+
'slot': ''
|
341 |
+
})
|
342 |
+
elif intent == 'Request':
|
343 |
+
assert value == '' and slot != 'none'
|
344 |
+
converted_da['binary'].append({
|
345 |
+
'intent': intent,
|
346 |
+
'domain': domain,
|
347 |
+
'slot': slot
|
348 |
+
})
|
349 |
+
elif '酒店设施' in slot:
|
350 |
+
converted_da['binary'].append({
|
351 |
+
'intent': intent,
|
352 |
+
'domain': domain,
|
353 |
+
'slot': f"{slot}-{value}"
|
354 |
+
})
|
355 |
+
elif intent == 'Select':
|
356 |
+
assert slot == '源领域'
|
357 |
+
converted_da['binary'].append({
|
358 |
+
'intent': intent,
|
359 |
+
'domain': domain,
|
360 |
+
'slot': f"{slot}-{value}"
|
361 |
+
})
|
362 |
+
elif slot in ['酒店类型', '车型', '车牌']:
|
363 |
+
assert intent in ['Inform', 'Recommend']
|
364 |
+
assert slot != 'none' and value != 'none'
|
365 |
+
converted_da['categorical'].append({
|
366 |
+
'intent': intent,
|
367 |
+
'domain': domain,
|
368 |
+
'slot': slot,
|
369 |
+
'value': value
|
370 |
+
})
|
371 |
+
else:
|
372 |
+
assert intent in ['Inform', 'Recommend']
|
373 |
+
assert slot != 'none' and value != 'none'
|
374 |
+
matches = utt.count(value)
|
375 |
+
if matches == 1:
|
376 |
+
start = utt.index(value)
|
377 |
+
end = start + len(value)
|
378 |
+
|
379 |
+
converted_da['non-categorical'].append({
|
380 |
+
'intent': intent,
|
381 |
+
'domain': domain,
|
382 |
+
'slot': slot,
|
383 |
+
'value': value,
|
384 |
+
'start': start,
|
385 |
+
'end': end
|
386 |
+
})
|
387 |
+
cnt_domain_slot['have span'] += 1
|
388 |
+
else:
|
389 |
+
# can not find span
|
390 |
+
converted_da['non-categorical'].append({
|
391 |
+
'intent': intent,
|
392 |
+
'domain': domain,
|
393 |
+
'slot': slot,
|
394 |
+
'value': value
|
395 |
+
})
|
396 |
+
cnt_domain_slot['no span'] += 1
|
397 |
+
# cnt_domain_slot.setdefault(f'{domain}-{slot}', set())
|
398 |
+
# cnt_domain_slot[f'{domain}-{slot}'].add(value)
|
399 |
+
|
400 |
+
return converted_da
|
401 |
+
|
402 |
+
def transform_user_state(user_state):
|
403 |
+
goal = []
|
404 |
+
for subgoal in user_state:
|
405 |
+
gid, domain, slot, value, mentioned = subgoal
|
406 |
+
if len(value) != 0:
|
407 |
+
t = 'inform'
|
408 |
+
else:
|
409 |
+
t = 'request'
|
410 |
+
if len(goal) < gid:
|
411 |
+
goal.append({domain: {'inform': {}, 'request': {}}})
|
412 |
+
goal[gid-1][domain][t][slot] = [value, 'mentioned' if mentioned else 'not mentioned']
|
413 |
+
return goal
|
414 |
+
|
415 |
+
|
416 |
+
def preprocess():
|
417 |
+
original_data_dir = '../../crosswoz'
|
418 |
+
new_data_dir = 'data'
|
419 |
+
|
420 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
421 |
+
for filename in os.listdir(os.path.join(original_data_dir,'database')):
|
422 |
+
copy2(f'{original_data_dir}/database/{filename}', new_data_dir)
|
423 |
+
|
424 |
+
global ontology
|
425 |
+
|
426 |
+
dataset = 'crosswoz'
|
427 |
+
splits = ['train', 'validation', 'test']
|
428 |
+
dialogues_by_split = {split: [] for split in splits}
|
429 |
+
for split in ['train', 'val', 'test']:
|
430 |
+
data = json.load(ZipFile(os.path.join(original_data_dir, f'{split}.json.zip'), 'r').open(f'{split}.json'))
|
431 |
+
if split == 'val':
|
432 |
+
split = 'validation'
|
433 |
+
|
434 |
+
for ori_dialog_id, ori_dialog in data.items():
|
435 |
+
if ori_dialog_id in ['10550', '11724']:
|
436 |
+
# skip error dialog
|
437 |
+
continue
|
438 |
+
dialogue_id = f'{dataset}-{split}-{len(dialogues_by_split[split])}'
|
439 |
+
|
440 |
+
# get user goal and involved domains
|
441 |
+
goal = {'inform': {}, 'request': {}}
|
442 |
+
goal["description"] = '\n'.join(ori_dialog["task description"])
|
443 |
+
cur_domains = [x[1] for i, x in enumerate(ori_dialog['goal']) if i == 0 or ori_dialog['goal'][i-1][1] != x[1]]
|
444 |
+
|
445 |
+
dialogue = {
|
446 |
+
'dataset': dataset,
|
447 |
+
'data_split': split,
|
448 |
+
'dialogue_id': dialogue_id,
|
449 |
+
'original_id': ori_dialog_id,
|
450 |
+
'domains': cur_domains,
|
451 |
+
'goal': goal,
|
452 |
+
'user_state_init': transform_user_state(ori_dialog['goal']),
|
453 |
+
'type': ori_dialog['type'],
|
454 |
+
'turns': [],
|
455 |
+
'user_state_final': transform_user_state(ori_dialog['final_goal'])
|
456 |
+
}
|
457 |
+
|
458 |
+
for turn_id, turn in enumerate(ori_dialog['messages']):
|
459 |
+
if ori_dialog_id == '2660' and turn_id in [8,9]:
|
460 |
+
# skip error turns
|
461 |
+
continue
|
462 |
+
elif ori_dialog_id == '7467' and turn_id in [14,15]:
|
463 |
+
# skip error turns
|
464 |
+
continue
|
465 |
+
elif ori_dialog_id == '11726' and turn_id in [4,5]:
|
466 |
+
# skip error turns
|
467 |
+
continue
|
468 |
+
speaker = 'user' if turn['role'] == 'usr' else 'system'
|
469 |
+
utt = turn['content']
|
470 |
+
|
471 |
+
das = turn['dialog_act']
|
472 |
+
|
473 |
+
dialogue_acts = convert_da(das, utt)
|
474 |
+
|
475 |
+
dialogue['turns'].append({
|
476 |
+
'speaker': speaker,
|
477 |
+
'utterance': utt,
|
478 |
+
'utt_idx': len(dialogue['turns']),
|
479 |
+
'dialogue_acts': dialogue_acts,
|
480 |
+
})
|
481 |
+
|
482 |
+
# add to dialogue_acts dictionary in the ontology
|
483 |
+
for da_type in dialogue_acts:
|
484 |
+
das = dialogue_acts[da_type]
|
485 |
+
for da in das:
|
486 |
+
ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
|
487 |
+
ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])][speaker] = True
|
488 |
+
|
489 |
+
if speaker == 'user':
|
490 |
+
dialogue['turns'][-1]['user_state'] = transform_user_state(turn['user_state'])
|
491 |
+
else:
|
492 |
+
# add state to last user turn
|
493 |
+
belief_state = turn['sys_state_init']
|
494 |
+
for domain in belief_state:
|
495 |
+
belief_state[domain].pop('selectedResults')
|
496 |
+
dialogue['turns'][-2]['state'] = belief_state
|
497 |
+
db_query = turn['sys_state']
|
498 |
+
db_results = {}
|
499 |
+
for domain in list(db_query.keys()):
|
500 |
+
db_res = db_query[domain].pop('selectedResults')
|
501 |
+
if len(db_res) > 0:
|
502 |
+
db_results[domain] = [{'名称': x} for x in db_res]
|
503 |
+
else:
|
504 |
+
db_query.pop(domain)
|
505 |
+
dialogue['turns'][-1]['db_query'] = db_query
|
506 |
+
dialogue['turns'][-1]['db_results'] = db_results
|
507 |
+
dialogues_by_split[split].append(dialogue)
|
508 |
+
pprint(cnt_domain_slot.most_common())
|
509 |
+
dialogues = []
|
510 |
+
for split in splits:
|
511 |
+
dialogues += dialogues_by_split[split]
|
512 |
+
for da_type in ontology['dialogue_acts']:
|
513 |
+
ontology["dialogue_acts"][da_type] = sorted([str(
|
514 |
+
{'user': speakers.get('user', False), 'system': speakers.get('system', False), 'intent': da[0],
|
515 |
+
'domain': da[1], 'slot': da[2]}) for da, speakers in ontology["dialogue_acts"][da_type].items()])
|
516 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
517 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
518 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
519 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
520 |
+
for filename in os.listdir(new_data_dir):
|
521 |
+
zf.write(f'{new_data_dir}/{filename}')
|
522 |
+
rmtree(new_data_dir)
|
523 |
+
return dialogues, ontology
|
524 |
+
|
525 |
+
|
526 |
+
def fix_entity_booked_info(entity_booked_dict, booked):
|
527 |
+
for domain in entity_booked_dict:
|
528 |
+
if not entity_booked_dict[domain] and booked[domain]:
|
529 |
+
entity_booked_dict[domain] = True
|
530 |
+
booked[domain] = []
|
531 |
+
return entity_booked_dict, booked
|
532 |
+
|
533 |
+
|
534 |
+
if __name__ == '__main__':
|
535 |
+
preprocess()
|
shuffled_dial_ids.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|