Dataset Card for "cardiffnlp/tweet_topic_multi"

Dataset Summary

This is the official repository of TweetTopic ("Twitter Topic Classification , COLING main conference 2022"), a topic classification dataset on Twitter with 19 labels. Each instance of TweetTopic comes with a timestamp which distributes from September 2019 to August 2021. See cardiffnlp/tweet_topic_single for single label version of TweetTopic. The tweet collection used in TweetTopic is same as what used in TweetNER7. The dataset is integrated in TweetNLP too.

Preprocessing

We pre-process tweets before the annotation to normalize some artifacts, converting URLs into a special token {{URL}} and non-verified usernames into {{USERNAME}}. For verified usernames, we replace its display name (or account name) with symbols {@}. For example, a tweet

Get the all-analog Classic Vinyl Edition
of "Takin' Off" Album from @herbiehancock
via @bluenoterecords link below: 
http://bluenote.lnk.to/AlbumOfTheWeek

is transformed into the following text.

Get the all-analog Classic Vinyl Edition
of "Takin' Off" Album from {@herbiehancock@}
via {@bluenoterecords@} link below: {{URL}}

A simple function to format tweet follows below.

import re
from urlextract import URLExtract
extractor = URLExtract()

def format_tweet(tweet):
    # mask web urls
    urls = extractor.find_urls(tweet)
    for url in urls:
        tweet = tweet.replace(url, "{{URL}}")
    # format twitter account
    tweet = re.sub(r"\b(\s*)(@[\S]+)\b", r'\1{\2@}', tweet)
    return tweet

target = """Get the all-analog Classic Vinyl Edition of "Takin' Off" Album from @herbiehancock via @bluenoterecords link below: http://bluenote.lnk.to/AlbumOfTheWeek"""
target_format = format_tweet(target)
print(target_format)
'Get the all-analog Classic Vinyl Edition of "Takin\' Off" Album from {@herbiehancock@} via {@bluenoterecords@} link below: {{URL}}'

Data Splits

split	number of texts	description
test_2020	573	test dataset from September 2019 to August 2020
test_2021	1679	test dataset from September 2020 to August 2021
train_2020	4585	training dataset from September 2019 to August 2020
train_2021	1505	training dataset from September 2020 to August 2021
train_all	6090	combined training dataset of train_2020 and train_2021
validation_2020	573	validation dataset from September 2019 to August 2020
validation_2021	188	validation dataset from September 2020 to August 2021
train_random	4564	randomly sampled training dataset with the same size as train_2020 from train_all
validation_random	573	randomly sampled training dataset with the same size as validation_2020 from validation_all
test_coling2022_random	5536	random split used in the COLING 2022 paper
train_coling2022_random	5731	random split used in the COLING 2022 paper
test_coling2022	5536	temporal split used in the COLING 2022 paper
train_coling2022	5731	temporal split used in the COLING 2022 paper

For the temporal-shift setting, model should be trained on train_2020 with validation_2020 and evaluate on test_2021. In general, model would be trained on train_all, the most representative training set with validation_2021 and evaluate on test_2021.

IMPORTANT NOTE: To get a result that is comparable with the results of the COLING 2022 Tweet Topic paper, please use train_coling2022 and test_coling2022 for temporal-shift, and train_coling2022_random and test_coling2022_random fir random split (the coling2022 split does not have validation set).

Models

model	training data	F1	F1 (macro)	Accuracy
cardiffnlp/roberta-large-tweet-topic-multi-all	all (2020 + 2021)	0.763104	0.620257	0.536629
cardiffnlp/roberta-base-tweet-topic-multi-all	all (2020 + 2021)	0.751814	0.600782	0.531864
cardiffnlp/twitter-roberta-base-2019-90m-tweet-topic-multi-all	all (2020 + 2021)	0.762513	0.603533	0.547945
cardiffnlp/twitter-roberta-base-dec2020-tweet-topic-multi-all	all (2020 + 2021)	0.759917	0.59901	0.536033
cardiffnlp/twitter-roberta-base-dec2021-tweet-topic-multi-all	all (2020 + 2021)	0.764767	0.618702	0.548541
cardiffnlp/roberta-large-tweet-topic-multi-2020	2020 only	0.732366	0.579456	0.493746
cardiffnlp/roberta-base-tweet-topic-multi-2020	2020 only	0.725229	0.561261	0.499107
cardiffnlp/twitter-roberta-base-2019-90m-tweet-topic-multi-2020	2020 only	0.73671	0.565624	0.513401
cardiffnlp/twitter-roberta-base-dec2020-tweet-topic-multi-2020	2020 only	0.729446	0.534799	0.50268
cardiffnlp/twitter-roberta-base-dec2021-tweet-topic-multi-2020	2020 only	0.731106	0.532141	0.509827

Model fine-tuning script can be found here.

Dataset Structure

Data Instances

An example of train looks as follows.

{
    "date": "2021-03-07",
    "text": "The latest The Movie theater Daily! {{URL}} Thanks to {{USERNAME}} {{USERNAME}} {{USERNAME}} #lunchtimeread #amc1000",
    "id": "1368464923370676231",
    "label": [0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
    "label_name": ["film_tv_&_video"]
}

Labels

0: arts_&_culture	5: fashion_&_style	10: learning_&_educational	15: science_&_technology
1: business_&_entrepreneurs	6: film_tv_&_video	11: music	16: sports
2: celebrity_&_pop_culture	7: fitness_&_health	12: news_&_social_concern	17: travel_&_adventure
3: diaries_&_daily_life	8: food_&_dining	13: other_hobbies	18: youth_&_student_life
4: family	9: gaming	14: relationships

Annotation instructions can be found here.

The label2id dictionary can be found here.

Citation Information

@inproceedings{dimosthenis-etal-2022-twitter,
    title = "{T}witter {T}opic {C}lassification",
    author = "Antypas, Dimosthenis  and
    Ushio, Asahi  and
    Camacho-Collados, Jose  and
    Neves, Leonardo  and
    Silva, Vitor  and
    Barbieri, Francesco",
    booktitle = "Proceedings of the 29th International Conference on Computational Linguistics",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "International Committee on Computational Linguistics"
}