from zipfile import ZipFile, ZIP_DEFLATED import json import os import copy import zipfile from tqdm import tqdm import re from collections import Counter from shutil import rmtree from convlab.util.file_util import read_zipped_json, write_zipped_json from pprint import pprint import random descriptions = { "flights": { "flights": "find a round trip or multi-city flights", "type": "type of the flight", "destination1": "the first destination city of the trip", "destination2": "the second destination city of the trip", "origin": "the origin city of the trip", "date.depart_origin": "date of departure from origin", "date.depart_intermediate": "date of departure from intermediate", "date.return": "date of return", "time_of_day": "time of the flight", "seating_class": "seat type (first class, business class, economy class, etc.", "seat_location": "location of the seat", "stops": "non-stop, layovers, etc.", "price_range": "price range of the flight", "num.pax": "number of people", "luggage": "luggage information", "total_fare": "total cost of the trip", "other_description": "other description of the flight", "from": "departure of the flight", "to": "destination of the flight", "airline": "airline of the flight", "flight_number": "the number of the flight", "date": "date of the flight", "from.time": "departure time of the flight", "to.time": "arrival time of the flight", "stops.location": "location of the stop", "fare": "cost of the flight", }, "food-ordering": { "food-ordering": "order take-out for a particular cuisine choice", "name.item": "name of the item", "other_description.item": "other description of the item", "type.retrieval": "type of the retrieval method", "total_price": "total price", "time.pickup": "pick up time", "num.people": "number of people", "name.restaurant": "name of the restaurant", "type.food": "type of food", "type.meal": "type of meal", "location.restaurant": "location of the restaurant", "rating.restaurant": "rating of the restaurant", "price_range": "price range of the food", }, "hotels": { "hotels": "find a hotel using typical preferences", "name.hotel": "name of the hotel", "location.hotel": "location of the hotel", "sub_location.hotel": "rough location of the hotel", "star_rating": "star rating of the hotel", "customer_rating": "customer rating of the hotel", "customer_review": "customer review of the hotel", "price_range": "price range of the hotel", "amenity": "amenity of the hotel", "num.beds": "number of beds to book", "type.bed": "type of the bed", "num.rooms": "number of rooms to book", "check-in_date": "check-in date", "check-out_date": "check-out date", "date_range": "date range of the reservation", "num.guests": "number of guests", "type.room": "type of the room", "price_per_night": "price per night", "total_fare": "total fare", "location": "location of the hotel", "other_request": "other request", "other_detail": "other detail", }, "movies": { "movies": "find a movie to watch in theaters or using a streaming service at home", "name.movie": "name of the movie", "genre": "genre of the movie", "name.theater": "name of the theater", "location.theater": "location of the theater", "time.start": "start time of the movie", "time.end": "end time of the movie", "price.ticket": "price of the ticket", "price.streaming": "price of the streaming", "type.screening": "type of the screening", "audience_rating": "audience rating", "critic_rating": "critic rating", "movie_rating": "film rating", "release_date": "release date of the movie", "runtime": "running time of the movie", "real_person": "name of actors, directors, etc.", "character": "name of character in the movie", "streaming_service": "streaming service that provide the movie", "num.tickets": "number of tickets", "seating": "type of seating", "other_description": "other description about the movie", "synopsis": "synopsis of the movie", }, "music": { "music": "find several tracks to play and then comment on each one", "name.track": "name of the track", "name.artist": "name of the artist", "name.album": "name of the album", "name.genre": "music genre", "type.music": "rough type of the music", "describes_track": "description of a track to find", "describes_artist": "description of a artist to find", "describes_album": "description of an album to find", "describes_genre": "description of a genre to find", "describes_type.music": "description of the music type", "technical_difficulty": "there is a technical difficulty", }, "restaurant-search": { "restaurant-search": "ask for recommendations for a particular type of cuisine", "name.restaurant": "name of the restaurant", "location": "location of the restaurant", "sub-location": "rough location of the restaurant", "type.food": "the cuisine of the restaurant", "menu_item": "item in the menu", "type.meal": "type of meal", "rating": "rating of the restaurant", "price_range": "price range of the restaurant", "business_hours": "business hours of the restaurant", "name.reservation": "name of the person who make the reservation", "num.guests": "number of guests", "time.reservation": "time of the reservation", "date.reservation": "date of the reservation", "type.seating": "type of the seating", "other_description": "other description of the restaurant", "phone": "phone number of the restaurant", }, "sports": { "sports": "discuss facts and stats about players, teams, games, etc. in EPL, MLB, MLS, NBA, NFL", "name.team": "name of the team", "record.team": "record of the team (number of wins and losses)", "record.games_ahead": "number of games ahead", "record.games_back": "number of games behind", "place.team": "ranking of the team", "result.match": "result of the match", "score.match": "score of the match", "date.match": "date of the match", "day.match": "day of the match", "time.match": "time of the match", "name.player": "name of the player", "position.player": "position of the player", "record.player": "record of the player", "name.non_player": "name of non-palyer such as the manager, coach", "venue": "venue of the match take place", "other_description.person": "other description of the person", "other_description.team": "other description of the team", "other_description.match": "other description of the match", } } anno2slot = { "flights": { "date.depart": "date.depart_origin", # rename "date.intermediate": "date.depart_intermediate", # rename "flight_booked": False, # transform to binary dialog act }, "food-ordering": { "name.person": None, # no sample, ignore "phone.restaurant": None, # no sample, ignore "business_hours.restaurant": None, # no sample, ignore "official_description.restaurant": None, # 1 sample, ignore }, "hotels": { "hotel_booked": False, # transform to binary dialog act }, "movies": { "time.end.": "time.end", # rename "seating ticket_booking": "seating", # mixed in the original ontology "ticket_booking": False, # transform to binary dialog act "synopsis": False, # too long, 54 words in avg. transform to binary dialog act }, "music": {}, "restaurant-search": { "offical_description": False, # too long, 15 words in avg. transform to binary dialog act }, "sports": {} } def format_turns(ori_turns): # delete invalid turns and merge continuous turns new_turns = [] previous_speaker = None utt_idx = 0 for i, turn in enumerate(ori_turns): speaker = 'system' if turn['speaker'] == 'ASSISTANT' else 'user' turn['speaker'] = speaker if turn['text'] == '(deleted)': continue if not previous_speaker: # first turn assert speaker != previous_speaker if speaker != previous_speaker: # switch speaker previous_speaker = speaker new_turns.append(copy.deepcopy(turn)) utt_idx += 1 else: # continuous speaking of the same speaker last_turn = new_turns[-1] # skip repeated turn if turn['text'] in ori_turns[i-1]['text']: continue # merge continuous turns index_shift = len(last_turn['text']) + 1 last_turn['text'] += ' '+turn['text'] if 'segments' in turn: last_turn.setdefault('segments', []) for segment in turn['segments']: segment['start_index'] += index_shift segment['end_index'] += index_shift last_turn['segments'] += turn['segments'] return new_turns def preprocess(): original_data_dir = 'Taskmaster-master' new_data_dir = 'data' if not os.path.exists(original_data_dir): original_data_zip = 'master.zip' if not os.path.exists(original_data_zip): 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') else: archive = ZipFile(original_data_zip) archive.extractall() os.makedirs(new_data_dir, exist_ok=True) ontology = {'domains': {}, 'intents': { 'inform': {'description': 'inform the value of a slot or general information.'} }, 'state': {}, 'dialogue_acts': { "categorical": {}, "non-categorical": {}, "binary": {} }} global descriptions global anno2slot domains = ['flights', 'food-ordering', 'hotels', 'movies', 'music', 'restaurant-search', 'sports'] for domain in domains: domain_ontology = json.load(open(os.path.join(original_data_dir, f"TM-2-2020/ontology/{domain}.json"))) assert len(domain_ontology) == 1 ontology['domains'][domain] = {'description': descriptions[domain][domain], 'slots': {}} ontology['state'][domain] = {} for item in list(domain_ontology.values())[0]: for anno in item['annotations']: slot = anno.strip() if slot in anno2slot[domain]: if anno2slot[domain][slot] in [None, False]: continue else: slot = anno2slot[domain][slot] ontology['domains'][domain]['slots'][slot] = { 'description': descriptions[domain][slot], 'is_categorical': False, 'possible_values': [], } ontology['state'][domain][slot] = '' # add missing slots to the ontology for domain, slot in [('movies', 'price.streaming'), ('restaurant-search', 'phone')]: ontology['domains'][domain]['slots'][slot] = { 'description': descriptions[domain][slot], 'is_categorical': False, 'possible_values': [], } ontology['state'][domain][slot] = '' dataset = 'tm2' splits = ['train', 'validation', 'test'] dialogues_by_split = {split:[] for split in splits} for domain in domains: data = json.load(open(os.path.join(original_data_dir, f"TM-2-2020/data/{domain}.json"))) # random split, train:validation:test = 8:1:1 random.seed(42) dial_ids = list(range(len(data))) random.shuffle(dial_ids) dial_id2split = {} for dial_id in dial_ids[:int(0.8*len(dial_ids))]: dial_id2split[dial_id] = 'train' for dial_id in dial_ids[int(0.8*len(dial_ids)):int(0.9*len(dial_ids))]: dial_id2split[dial_id] = 'validation' for dial_id in dial_ids[int(0.9*len(dial_ids)):]: dial_id2split[dial_id] = 'test' for dial_id, d in tqdm(enumerate(data), desc='processing taskmaster-{}'.format(domain)): # delete empty dialogs and invalid dialogs if len(d['utterances']) == 0: continue if len(set([t['speaker'] for t in d['utterances']])) == 1: continue data_split = dial_id2split[dial_id] dialogue_id = f'{dataset}-{data_split}-{len(dialogues_by_split[data_split])}' cur_domains = [domain] dialogue = { 'dataset': dataset, 'data_split': data_split, 'dialogue_id': dialogue_id, 'original_id': d["conversation_id"], 'domains': cur_domains, 'turns': [] } turns = format_turns(d['utterances']) prev_state = {} prev_state.setdefault(domain, copy.deepcopy(ontology['state'][domain])) for utt_idx, uttr in enumerate(turns): speaker = uttr['speaker'] turn = { 'speaker': speaker, 'utterance': uttr['text'], 'utt_idx': utt_idx, 'dialogue_acts': { 'binary': [], 'categorical': [], 'non-categorical': [], }, } in_span = [0] * len(turn['utterance']) if 'segments' in uttr: # sort the span according to the length segments = sorted(uttr['segments'], key=lambda x: len(x['text'])) for segment in segments: # Each conversation was annotated by two workers. # only keep the first annotation for the span item = segment['annotations'][0] intent = 'inform' # default intent slot = item['name'].split('.', 1)[-1].strip() if slot in anno2slot[domain]: if anno2slot[domain][slot] is None: # skip continue elif anno2slot[domain][slot] is False: # binary dialog act turn['dialogue_acts']['binary'].append({ 'intent': intent, 'domain': domain, 'slot': slot, }) continue else: slot = anno2slot[domain][slot] assert slot in ontology['domains'][domain]['slots'], print(domain, [slot]) assert turn['utterance'][segment['start_index']:segment['end_index']] == segment['text'] # skip overlapped spans, keep the shortest one if sum(in_span[segment['start_index']: segment['end_index']]) > 0: continue else: in_span[segment['start_index']: segment['end_index']] = [1]*(segment['end_index']-segment['start_index']) turn['dialogue_acts']['non-categorical'].append({ 'intent': intent, 'domain': domain, 'slot': slot, 'value': segment['text'], 'start': segment['start_index'], 'end': segment['end_index'] }) turn['dialogue_acts']['non-categorical'] = sorted(turn['dialogue_acts']['non-categorical'], key=lambda x: x['start']) bdas = set() for da in turn['dialogue_acts']['binary']: da_tuple = (da['intent'], da['domain'], da['slot'],) bdas.add(da_tuple) turn['dialogue_acts']['binary'] = [{'intent':bda[0],'domain':bda[1],'slot':bda[2]} for bda in sorted(bdas)] # add to dialogue_acts dictionary in the ontology for da_type in turn['dialogue_acts']: das = turn['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 for da in turn['dialogue_acts']['non-categorical']: slot, value = da['slot'], da['value'] assert slot in prev_state[domain] prev_state[domain][slot] = value if speaker == 'user': turn['state'] = copy.deepcopy(prev_state) dialogue['turns'].append(turn) dialogues_by_split[data_split].append(dialogue) 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()]) dialogues = dialogues_by_split['train']+dialogues_by_split['validation']+dialogues_by_split['test'] 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(original_data_dir) rmtree(new_data_dir) return dialogues, ontology if __name__ == '__main__': preprocess()