Upload preprocess.py
Browse files- preprocess.py +413 -0
preprocess.py
ADDED
|
@@ -0,0 +1,413 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from zipfile import ZipFile, ZIP_DEFLATED
|
| 2 |
+
import json
|
| 3 |
+
import os
|
| 4 |
+
import copy
|
| 5 |
+
import zipfile
|
| 6 |
+
from tqdm import tqdm
|
| 7 |
+
import re
|
| 8 |
+
from collections import Counter
|
| 9 |
+
from shutil import rmtree
|
| 10 |
+
from convlab.util.file_util import read_zipped_json, write_zipped_json
|
| 11 |
+
from pprint import pprint
|
| 12 |
+
import random
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
descriptions = {
|
| 16 |
+
"flights": {
|
| 17 |
+
"flights": "find a round trip or multi-city flights",
|
| 18 |
+
"type": "type of the flight",
|
| 19 |
+
"destination1": "the first destination city of the trip",
|
| 20 |
+
"destination2": "the second destination city of the trip",
|
| 21 |
+
"origin": "the origin city of the trip",
|
| 22 |
+
"date.depart_origin": "date of departure from origin",
|
| 23 |
+
"date.depart_intermediate": "date of departure from intermediate",
|
| 24 |
+
"date.return": "date of return",
|
| 25 |
+
"time_of_day": "time of the flight",
|
| 26 |
+
"seating_class": "seat type (first class, business class, economy class, etc.",
|
| 27 |
+
"seat_location": "location of the seat",
|
| 28 |
+
"stops": "non-stop, layovers, etc.",
|
| 29 |
+
"price_range": "price range of the flight",
|
| 30 |
+
"num.pax": "number of people",
|
| 31 |
+
"luggage": "luggage information",
|
| 32 |
+
"total_fare": "total cost of the trip",
|
| 33 |
+
"other_description": "other description of the flight",
|
| 34 |
+
"from": "departure of the flight",
|
| 35 |
+
"to": "destination of the flight",
|
| 36 |
+
"airline": "airline of the flight",
|
| 37 |
+
"flight_number": "the number of the flight",
|
| 38 |
+
"date": "date of the flight",
|
| 39 |
+
"from.time": "departure time of the flight",
|
| 40 |
+
"to.time": "arrival time of the flight",
|
| 41 |
+
"stops.location": "location of the stop",
|
| 42 |
+
"fare": "cost of the flight",
|
| 43 |
+
},
|
| 44 |
+
"food-ordering": {
|
| 45 |
+
"food-ordering": "order take-out for a particular cuisine choice",
|
| 46 |
+
"name.item": "name of the item",
|
| 47 |
+
"other_description.item": "other description of the item",
|
| 48 |
+
"type.retrieval": "type of the retrieval method",
|
| 49 |
+
"total_price": "total price",
|
| 50 |
+
"time.pickup": "pick up time",
|
| 51 |
+
"num.people": "number of people",
|
| 52 |
+
"name.restaurant": "name of the restaurant",
|
| 53 |
+
"type.food": "type of food",
|
| 54 |
+
"type.meal": "type of meal",
|
| 55 |
+
"location.restaurant": "location of the restaurant",
|
| 56 |
+
"rating.restaurant": "rating of the restaurant",
|
| 57 |
+
"price_range": "price range of the food",
|
| 58 |
+
},
|
| 59 |
+
"hotels": {
|
| 60 |
+
"hotels": "find a hotel using typical preferences",
|
| 61 |
+
"name.hotel": "name of the hotel",
|
| 62 |
+
"location.hotel": "location of the hotel",
|
| 63 |
+
"sub_location.hotel": "rough location of the hotel",
|
| 64 |
+
"star_rating": "star rating of the hotel",
|
| 65 |
+
"customer_rating": "customer rating of the hotel",
|
| 66 |
+
"customer_review": "customer review of the hotel",
|
| 67 |
+
"price_range": "price range of the hotel",
|
| 68 |
+
"amenity": "amenity of the hotel",
|
| 69 |
+
"num.beds": "number of beds to book",
|
| 70 |
+
"type.bed": "type of the bed",
|
| 71 |
+
"num.rooms": "number of rooms to book",
|
| 72 |
+
"check-in_date": "check-in date",
|
| 73 |
+
"check-out_date": "check-out date",
|
| 74 |
+
"date_range": "date range of the reservation",
|
| 75 |
+
"num.guests": "number of guests",
|
| 76 |
+
"type.room": "type of the room",
|
| 77 |
+
"price_per_night": "price per night",
|
| 78 |
+
"total_fare": "total fare",
|
| 79 |
+
"location": "location of the hotel",
|
| 80 |
+
"other_request": "other request",
|
| 81 |
+
"other_detail": "other detail",
|
| 82 |
+
},
|
| 83 |
+
"movies": {
|
| 84 |
+
"movies": "find a movie to watch in theaters or using a streaming service at home",
|
| 85 |
+
"name.movie": "name of the movie",
|
| 86 |
+
"genre": "genre of the movie",
|
| 87 |
+
"name.theater": "name of the theater",
|
| 88 |
+
"location.theater": "location of the theater",
|
| 89 |
+
"time.start": "start time of the movie",
|
| 90 |
+
"time.end": "end time of the movie",
|
| 91 |
+
"price.ticket": "price of the ticket",
|
| 92 |
+
"price.streaming": "price of the streaming",
|
| 93 |
+
"type.screening": "type of the screening",
|
| 94 |
+
"audience_rating": "audience rating",
|
| 95 |
+
"critic_rating": "critic rating",
|
| 96 |
+
"movie_rating": "film rating",
|
| 97 |
+
"release_date": "release date of the movie",
|
| 98 |
+
"runtime": "running time of the movie",
|
| 99 |
+
"real_person": "name of actors, directors, etc.",
|
| 100 |
+
"character": "name of character in the movie",
|
| 101 |
+
"streaming_service": "streaming service that provide the movie",
|
| 102 |
+
"num.tickets": "number of tickets",
|
| 103 |
+
"seating": "type of seating",
|
| 104 |
+
"other_description": "other description about the movie",
|
| 105 |
+
"synopsis": "synopsis of the movie",
|
| 106 |
+
},
|
| 107 |
+
"music": {
|
| 108 |
+
"music": "find several tracks to play and then comment on each one",
|
| 109 |
+
"name.track": "name of the track",
|
| 110 |
+
"name.artist": "name of the artist",
|
| 111 |
+
"name.album": "name of the album",
|
| 112 |
+
"name.genre": "music genre",
|
| 113 |
+
"type.music": "rough type of the music",
|
| 114 |
+
"describes_track": "description of a track to find",
|
| 115 |
+
"describes_artist": "description of a artist to find",
|
| 116 |
+
"describes_album": "description of an album to find",
|
| 117 |
+
"describes_genre": "description of a genre to find",
|
| 118 |
+
"describes_type.music": "description of the music type",
|
| 119 |
+
"technical_difficulty": "there is a technical difficulty",
|
| 120 |
+
},
|
| 121 |
+
"restaurant-search": {
|
| 122 |
+
"restaurant-search": "ask for recommendations for a particular type of cuisine",
|
| 123 |
+
"name.restaurant": "name of the restaurant",
|
| 124 |
+
"location": "location of the restaurant",
|
| 125 |
+
"sub-location": "rough location of the restaurant",
|
| 126 |
+
"type.food": "the cuisine of the restaurant",
|
| 127 |
+
"menu_item": "item in the menu",
|
| 128 |
+
"type.meal": "type of meal",
|
| 129 |
+
"rating": "rating of the restaurant",
|
| 130 |
+
"price_range": "price range of the restaurant",
|
| 131 |
+
"business_hours": "business hours of the restaurant",
|
| 132 |
+
"name.reservation": "name of the person who make the reservation",
|
| 133 |
+
"num.guests": "number of guests",
|
| 134 |
+
"time.reservation": "time of the reservation",
|
| 135 |
+
"date.reservation": "date of the reservation",
|
| 136 |
+
"type.seating": "type of the seating",
|
| 137 |
+
"other_description": "other description of the restaurant",
|
| 138 |
+
"phone": "phone number of the restaurant",
|
| 139 |
+
},
|
| 140 |
+
"sports": {
|
| 141 |
+
"sports": "discuss facts and stats about players, teams, games, etc. in EPL, MLB, MLS, NBA, NFL",
|
| 142 |
+
"name.team": "name of the team",
|
| 143 |
+
"record.team": "record of the team (number of wins and losses)",
|
| 144 |
+
"record.games_ahead": "number of games ahead",
|
| 145 |
+
"record.games_back": "number of games behind",
|
| 146 |
+
"place.team": "ranking of the team",
|
| 147 |
+
"result.match": "result of the match",
|
| 148 |
+
"score.match": "score of the match",
|
| 149 |
+
"date.match": "date of the match",
|
| 150 |
+
"day.match": "day of the match",
|
| 151 |
+
"time.match": "time of the match",
|
| 152 |
+
"name.player": "name of the player",
|
| 153 |
+
"position.player": "position of the player",
|
| 154 |
+
"record.player": "record of the player",
|
| 155 |
+
"name.non_player": "name of non-palyer such as the manager, coach",
|
| 156 |
+
"venue": "venue of the match take place",
|
| 157 |
+
"other_description.person": "other description of the person",
|
| 158 |
+
"other_description.team": "other description of the team",
|
| 159 |
+
"other_description.match": "other description of the match",
|
| 160 |
+
}
|
| 161 |
+
}
|
| 162 |
+
|
| 163 |
+
anno2slot = {
|
| 164 |
+
"flights": {
|
| 165 |
+
"date.depart": "date.depart_origin", # rename
|
| 166 |
+
"date.intermediate": "date.depart_intermediate", # rename
|
| 167 |
+
"flight_booked": False, # transform to binary dialog act
|
| 168 |
+
},
|
| 169 |
+
"food-ordering": {
|
| 170 |
+
"name.person": None, # no sample, ignore
|
| 171 |
+
"phone.restaurant": None, # no sample, ignore
|
| 172 |
+
"business_hours.restaurant": None, # no sample, ignore
|
| 173 |
+
"official_description.restaurant": None, # 1 sample, ignore
|
| 174 |
+
},
|
| 175 |
+
"hotels": {
|
| 176 |
+
"hotel_booked": False, # transform to binary dialog act
|
| 177 |
+
},
|
| 178 |
+
"movies": {
|
| 179 |
+
"time.end.": "time.end", # rename
|
| 180 |
+
"seating ticket_booking": "seating", # mixed in the original ontology
|
| 181 |
+
"ticket_booking": False, # transform to binary dialog act
|
| 182 |
+
"synopsis": False, # too long, 54 words in avg. transform to binary dialog act
|
| 183 |
+
},
|
| 184 |
+
"music": {},
|
| 185 |
+
"restaurant-search": {
|
| 186 |
+
"offical_description": False, # too long, 15 words in avg. transform to binary dialog act
|
| 187 |
+
},
|
| 188 |
+
"sports": {}
|
| 189 |
+
}
|
| 190 |
+
|
| 191 |
+
|
| 192 |
+
def format_turns(ori_turns):
|
| 193 |
+
# delete invalid turns and merge continuous turns
|
| 194 |
+
new_turns = []
|
| 195 |
+
previous_speaker = None
|
| 196 |
+
utt_idx = 0
|
| 197 |
+
for i, turn in enumerate(ori_turns):
|
| 198 |
+
speaker = 'system' if turn['speaker'] == 'ASSISTANT' else 'user'
|
| 199 |
+
turn['speaker'] = speaker
|
| 200 |
+
if turn['text'] == '(deleted)':
|
| 201 |
+
continue
|
| 202 |
+
if not previous_speaker:
|
| 203 |
+
# first turn
|
| 204 |
+
assert speaker != previous_speaker
|
| 205 |
+
if speaker != previous_speaker:
|
| 206 |
+
# switch speaker
|
| 207 |
+
previous_speaker = speaker
|
| 208 |
+
new_turns.append(copy.deepcopy(turn))
|
| 209 |
+
utt_idx += 1
|
| 210 |
+
else:
|
| 211 |
+
# continuous speaking of the same speaker
|
| 212 |
+
last_turn = new_turns[-1]
|
| 213 |
+
# skip repeated turn
|
| 214 |
+
if turn['text'] in ori_turns[i-1]['text']:
|
| 215 |
+
continue
|
| 216 |
+
# merge continuous turns
|
| 217 |
+
index_shift = len(last_turn['text']) + 1
|
| 218 |
+
last_turn['text'] += ' '+turn['text']
|
| 219 |
+
if 'segments' in turn:
|
| 220 |
+
last_turn.setdefault('segments', [])
|
| 221 |
+
for segment in turn['segments']:
|
| 222 |
+
segment['start_index'] += index_shift
|
| 223 |
+
segment['end_index'] += index_shift
|
| 224 |
+
last_turn['segments'] += turn['segments']
|
| 225 |
+
return new_turns
|
| 226 |
+
|
| 227 |
+
|
| 228 |
+
def preprocess():
|
| 229 |
+
original_data_dir = 'Taskmaster-master'
|
| 230 |
+
new_data_dir = 'data'
|
| 231 |
+
|
| 232 |
+
if not os.path.exists(original_data_dir):
|
| 233 |
+
original_data_zip = 'master.zip'
|
| 234 |
+
if not os.path.exists(original_data_zip):
|
| 235 |
+
raise FileNotFoundError(f'cannot find original data {original_data_zip} in tm2/, should manually download master.zip from https://github.com/google-research-datasets/Taskmaster/archive/refs/heads/master.zip')
|
| 236 |
+
else:
|
| 237 |
+
archive = ZipFile(original_data_zip)
|
| 238 |
+
archive.extractall()
|
| 239 |
+
|
| 240 |
+
os.makedirs(new_data_dir, exist_ok=True)
|
| 241 |
+
|
| 242 |
+
ontology = {'domains': {},
|
| 243 |
+
'intents': {
|
| 244 |
+
'inform': {'description': 'inform the value of a slot or general information.'}
|
| 245 |
+
},
|
| 246 |
+
'state': {},
|
| 247 |
+
'dialogue_acts': {
|
| 248 |
+
"categorical": {},
|
| 249 |
+
"non-categorical": {},
|
| 250 |
+
"binary": {}
|
| 251 |
+
}}
|
| 252 |
+
global descriptions
|
| 253 |
+
global anno2slot
|
| 254 |
+
domains = ['flights', 'food-ordering', 'hotels', 'movies', 'music', 'restaurant-search', 'sports']
|
| 255 |
+
for domain in domains:
|
| 256 |
+
domain_ontology = json.load(open(os.path.join(original_data_dir, f"TM-2-2020/ontology/{domain}.json")))
|
| 257 |
+
assert len(domain_ontology) == 1
|
| 258 |
+
ontology['domains'][domain] = {'description': descriptions[domain][domain], 'slots': {}}
|
| 259 |
+
ontology['state'][domain] = {}
|
| 260 |
+
for item in list(domain_ontology.values())[0]:
|
| 261 |
+
for anno in item['annotations']:
|
| 262 |
+
slot = anno.strip()
|
| 263 |
+
if slot in anno2slot[domain]:
|
| 264 |
+
if anno2slot[domain][slot] in [None, False]:
|
| 265 |
+
continue
|
| 266 |
+
else:
|
| 267 |
+
slot = anno2slot[domain][slot]
|
| 268 |
+
ontology['domains'][domain]['slots'][slot] = {
|
| 269 |
+
'description': descriptions[domain][slot],
|
| 270 |
+
'is_categorical': False,
|
| 271 |
+
'possible_values': [],
|
| 272 |
+
}
|
| 273 |
+
ontology['state'][domain][slot] = ''
|
| 274 |
+
# add missing slots to the ontology
|
| 275 |
+
for domain, slot in [('movies', 'price.streaming'), ('restaurant-search', 'phone')]:
|
| 276 |
+
ontology['domains'][domain]['slots'][slot] = {
|
| 277 |
+
'description': descriptions[domain][slot],
|
| 278 |
+
'is_categorical': False,
|
| 279 |
+
'possible_values': [],
|
| 280 |
+
}
|
| 281 |
+
ontology['state'][domain][slot] = ''
|
| 282 |
+
|
| 283 |
+
dataset = 'tm2'
|
| 284 |
+
splits = ['train', 'validation', 'test']
|
| 285 |
+
dialogues_by_split = {split:[] for split in splits}
|
| 286 |
+
for domain in domains:
|
| 287 |
+
data = json.load(open(os.path.join(original_data_dir, f"TM-2-2020/data/{domain}.json")))
|
| 288 |
+
# random split, train:validation:test = 8:1:1
|
| 289 |
+
random.seed(42)
|
| 290 |
+
dial_ids = list(range(len(data)))
|
| 291 |
+
random.shuffle(dial_ids)
|
| 292 |
+
dial_id2split = {}
|
| 293 |
+
for dial_id in dial_ids[:int(0.8*len(dial_ids))]:
|
| 294 |
+
dial_id2split[dial_id] = 'train'
|
| 295 |
+
for dial_id in dial_ids[int(0.8*len(dial_ids)):int(0.9*len(dial_ids))]:
|
| 296 |
+
dial_id2split[dial_id] = 'validation'
|
| 297 |
+
for dial_id in dial_ids[int(0.9*len(dial_ids)):]:
|
| 298 |
+
dial_id2split[dial_id] = 'test'
|
| 299 |
+
|
| 300 |
+
for dial_id, d in tqdm(enumerate(data), desc='processing taskmaster-{}'.format(domain)):
|
| 301 |
+
# delete empty dialogs and invalid dialogs
|
| 302 |
+
if len(d['utterances']) == 0:
|
| 303 |
+
continue
|
| 304 |
+
if len(set([t['speaker'] for t in d['utterances']])) == 1:
|
| 305 |
+
continue
|
| 306 |
+
data_split = dial_id2split[dial_id]
|
| 307 |
+
dialogue_id = f'{dataset}-{data_split}-{len(dialogues_by_split[data_split])}'
|
| 308 |
+
cur_domains = [domain]
|
| 309 |
+
dialogue = {
|
| 310 |
+
'dataset': dataset,
|
| 311 |
+
'data_split': data_split,
|
| 312 |
+
'dialogue_id': dialogue_id,
|
| 313 |
+
'original_id': d["conversation_id"],
|
| 314 |
+
'domains': cur_domains,
|
| 315 |
+
'turns': []
|
| 316 |
+
}
|
| 317 |
+
turns = format_turns(d['utterances'])
|
| 318 |
+
prev_state = {}
|
| 319 |
+
prev_state.setdefault(domain, copy.deepcopy(ontology['state'][domain]))
|
| 320 |
+
|
| 321 |
+
for utt_idx, uttr in enumerate(turns):
|
| 322 |
+
speaker = uttr['speaker']
|
| 323 |
+
turn = {
|
| 324 |
+
'speaker': speaker,
|
| 325 |
+
'utterance': uttr['text'],
|
| 326 |
+
'utt_idx': utt_idx,
|
| 327 |
+
'dialogue_acts': {
|
| 328 |
+
'binary': [],
|
| 329 |
+
'categorical': [],
|
| 330 |
+
'non-categorical': [],
|
| 331 |
+
},
|
| 332 |
+
}
|
| 333 |
+
in_span = [0] * len(turn['utterance'])
|
| 334 |
+
|
| 335 |
+
if 'segments' in uttr:
|
| 336 |
+
# sort the span according to the length
|
| 337 |
+
segments = sorted(uttr['segments'], key=lambda x: len(x['text']))
|
| 338 |
+
for segment in segments:
|
| 339 |
+
# Each conversation was annotated by two workers.
|
| 340 |
+
# only keep the first annotation for the span
|
| 341 |
+
item = segment['annotations'][0]
|
| 342 |
+
intent = 'inform' # default intent
|
| 343 |
+
slot = item['name'].split('.', 1)[-1].strip()
|
| 344 |
+
if slot in anno2slot[domain]:
|
| 345 |
+
if anno2slot[domain][slot] is None:
|
| 346 |
+
# skip
|
| 347 |
+
continue
|
| 348 |
+
elif anno2slot[domain][slot] is False:
|
| 349 |
+
# binary dialog act
|
| 350 |
+
turn['dialogue_acts']['binary'].append({
|
| 351 |
+
'intent': intent,
|
| 352 |
+
'domain': domain,
|
| 353 |
+
'slot': slot,
|
| 354 |
+
})
|
| 355 |
+
continue
|
| 356 |
+
else:
|
| 357 |
+
slot = anno2slot[domain][slot]
|
| 358 |
+
assert slot in ontology['domains'][domain]['slots'], print(domain, [slot])
|
| 359 |
+
assert turn['utterance'][segment['start_index']:segment['end_index']] == segment['text']
|
| 360 |
+
# skip overlapped spans, keep the shortest one
|
| 361 |
+
if sum(in_span[segment['start_index']: segment['end_index']]) > 0:
|
| 362 |
+
continue
|
| 363 |
+
else:
|
| 364 |
+
in_span[segment['start_index']: segment['end_index']] = [1]*(segment['end_index']-segment['start_index'])
|
| 365 |
+
turn['dialogue_acts']['non-categorical'].append({
|
| 366 |
+
'intent': intent,
|
| 367 |
+
'domain': domain,
|
| 368 |
+
'slot': slot,
|
| 369 |
+
'value': segment['text'],
|
| 370 |
+
'start': segment['start_index'],
|
| 371 |
+
'end': segment['end_index']
|
| 372 |
+
})
|
| 373 |
+
|
| 374 |
+
turn['dialogue_acts']['non-categorical'] = sorted(turn['dialogue_acts']['non-categorical'], key=lambda x: x['start'])
|
| 375 |
+
|
| 376 |
+
bdas = set()
|
| 377 |
+
for da in turn['dialogue_acts']['binary']:
|
| 378 |
+
da_tuple = (da['intent'], da['domain'], da['slot'],)
|
| 379 |
+
bdas.add(da_tuple)
|
| 380 |
+
turn['dialogue_acts']['binary'] = [{'intent':bda[0],'domain':bda[1],'slot':bda[2]} for bda in sorted(bdas)]
|
| 381 |
+
# add to dialogue_acts dictionary in the ontology
|
| 382 |
+
for da_type in turn['dialogue_acts']:
|
| 383 |
+
das = turn['dialogue_acts'][da_type]
|
| 384 |
+
for da in das:
|
| 385 |
+
ontology["dialogue_acts"][da_type].setdefault((da['intent'], da['domain'], da['slot']), {})
|
| 386 |
+
ontology["dialogue_acts"][da_type][(da['intent'], da['domain'], da['slot'])][speaker] = True
|
| 387 |
+
|
| 388 |
+
for da in turn['dialogue_acts']['non-categorical']:
|
| 389 |
+
slot, value = da['slot'], da['value']
|
| 390 |
+
assert slot in prev_state[domain]
|
| 391 |
+
prev_state[domain][slot] = value
|
| 392 |
+
|
| 393 |
+
if speaker == 'user':
|
| 394 |
+
turn['state'] = copy.deepcopy(prev_state)
|
| 395 |
+
|
| 396 |
+
dialogue['turns'].append(turn)
|
| 397 |
+
dialogues_by_split[data_split].append(dialogue)
|
| 398 |
+
|
| 399 |
+
for da_type in ontology['dialogue_acts']:
|
| 400 |
+
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()])
|
| 401 |
+
dialogues = dialogues_by_split['train']+dialogues_by_split['validation']+dialogues_by_split['test']
|
| 402 |
+
json.dump(dialogues[:10], open(f'dummy_data.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 403 |
+
json.dump(ontology, open(f'{new_data_dir}/ontology.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 404 |
+
json.dump(dialogues, open(f'{new_data_dir}/dialogues.json', 'w', encoding='utf-8'), indent=2, ensure_ascii=False)
|
| 405 |
+
with ZipFile('data.zip', 'w', ZIP_DEFLATED) as zf:
|
| 406 |
+
for filename in os.listdir(new_data_dir):
|
| 407 |
+
zf.write(f'{new_data_dir}/{filename}')
|
| 408 |
+
rmtree(original_data_dir)
|
| 409 |
+
rmtree(new_data_dir)
|
| 410 |
+
return dialogues, ontology
|
| 411 |
+
|
| 412 |
+
if __name__ == '__main__':
|
| 413 |
+
preprocess()
|