README.md

metadata

language:
  - en
license:
  - other
multilinguality:
  - monolingual
size_categories:
  - 1k<10K
task_categories:
  - token-classification
task_ids:
  - named-entity-recognition
pretty_name: TweetNER7

Dataset Card for "tner/tweetner7"

Dataset Description

• Repository: https://github.com/asahi417/tner/tree/master/examples/tweetner7_paper
• Paper: TBA
• Dataset: TweetNER7
• Domain: Twitter
• Number of Entity: 7

Dataset Summary

This is the official repository of TweetNER7 ("Named Entity Recognition in Twitter: A Dataset and Analysis on Short-Term Temporal Shifts, AACL main conference 2022"), an NER dataset on Twitter with 7 entity labels. Each instance of TweetNER7 comes with a timestamp which distributes from September 2019 to August 2021. The tweet collection used in TweetNER7 is same as what used in TweetTopic. The dataset is integrated in TweetNLP too.

• Entity Types: corperation, creative_work, event, group, location, product, person

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}}'

We ask annotators to ignore those special tokens but label the verified users' mentions.

Data Split

split	number of instances	description
train_2020	4616	training dataset from September 2019 to August 2020
train_2021	2495	training dataset from September 2020 to August 2021
train_all	7111	combined training dataset of train_2020 and train_2021
validation_2020	576	validation dataset from September 2019 to August 2020
validation_2021	310	validation dataset from September 2020 to August 2021
test_2020	576	test dataset from September 2019 to August 2020
test_2021	2807	test dataset from September 2020 to August 2021
train_random	4616	randomly sampled training dataset with the same size as train_2020 from train_all
validation_random	576	randomly sampled training dataset with the same size as validation_2020 from validation_all
extra_2020	87880	extra tweet without annotations from September 2019 to August 2020
extra_2021	93594	extra tweet without annotations from September 2020 to August 2021

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.

Dataset Structure

Data Instances

An example of train looks as follows.

{
    'tokens': ['Morning', '5km', 'run', 'with', '{{USERNAME}}', 'for', 'breast', 'cancer', 'awareness', '#', 'pinkoctober', '#', 'breastcancerawareness', '#', 'zalorafit', '#', 'zalorafitxbnwrc', '@', 'The', 'Central', 'Park', ',', 'Desa', 'Parkcity', '{{URL}}'],
    'tags': [14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 2, 14, 2, 14, 14, 14, 14, 14, 14, 4, 11, 11, 11, 11, 14],
    'id': '1183344337016381440',
    'date': '2019-10-13'
}

Label ID

The label2id dictionary can be found at here.

{
    "B-corporation": 0,
    "B-creative_work": 1,
    "B-event": 2,
    "B-group": 3,
    "B-location": 4,
    "B-person": 5,
    "B-product": 6,
    "I-corporation": 7,
    "I-creative_work": 8,
    "I-event": 9,
    "I-group": 10,
    "I-location": 11,
    "I-person": 12,
    "I-product": 13,
    "O": 14
}

Models

Main Models

Model (link)	Data	Language Model	Micro F1 (2021)	Macro F1 (2021)	F1 (2021)/corporation	F1 (2021)/creative_work	F1 (2021)/event	F1 (2021)/group	F1 (2021)/location	F1 (2021)/person	F1 (2021)/product	Micro F1 (2020)	Macro F1 (2020)	F1 (2020)/corporation	F1 (2020)/creative_work	F1 (2020)/event	F1 (2020)/group	F1 (2020)/location	F1 (2020)/person	F1 (2020)/product	Entity-Span F1 (2021)	Entity-Span F1 (2020)
tner/roberta-large-tweetner7-all	tweetner7	roberta-large	65.75	61.25	53.92	47.61	46.73	61.4	67.07	82.93	69.06	66.29	62.97	61.84	51.59	50.29	55.99	69.23	82.01	69.86	78.82	76.43
tner/roberta-base-tweetner7-all	tweetner7	roberta-base	65.16	60.81	51.74	46.64	46.73	60.71	68.33	83.77	67.77	65.32	61.66	61.94	48.94	45.14	56.58	68.94	82.75	67.33	78.93	75.23
tner/twitter-roberta-base-2019-90m-tweetner7-all	tweetner7	cardiffnlp/twitter-roberta-base-2019-90m	65.68	61	50.87	47.3	48.41	61.48	67.94	83.93	67.06	65.46	61.22	56.85	52.15	46.68	56.68	65.1	84.55	66.5	78.89	76.43
tner/twitter-roberta-base-dec2020-tweetner7-all	tweetner7	cardiffnlp/twitter-roberta-base-dec2020	65.26	60.7	51.53	47.6	46.69	60.93	66.89	83.87	67.38	65.44	61.39	56.76	55.06	46.24	55.52	64.26	84.87	67	78.68	75.87
tner/bertweet-large-tweetner7-all	tweetner7	cardiffnlp/twitter-roberta-base-dec2021vinai/bertweet-large	66.46	61.87	54.5	47.36	49.15	62.38	67.55	84.15	68.02	66.76	63.08	58.89	55.24	48.89	59.85	66.67	83.49	68.51	79.53	77.59
tner/bertweet-base-tweetner7-all	tweetner7	vinai/bertweet-base	65.36	60.52	52.51	46.54	48.06	60.33	65.67	84.08	66.46	65.74	61.61	57.22	54.1	48.55	57.35	64.57	84.16	65.36	78.99	76.91
tner/bert-large-tweetner7-all	tweetner7	bert-large	63.58	59	50.13	40.16	47	59.74	67.2	81.86	66.91	62.49	58.63	55.56	47.65	43.08	54.88	63.9	80.31	65.04	77.21	73.58
tner/bert-base-tweetner7-all	tweetner7	bert-base	62.3	57.59	51.41	38.86	45.81	56.61	62.65	81.97	65.8	62.1	57.74	56.55	41.52	45.04	54.23	60.53	81.86	64.49	76.62	72.98
tner/roberta-large-tweetner7-continuous	tweetner7	roberta-large	66.02	60.9	53.15	44.42	48.79	61.15	67.41	84.72	66.63	66.26	62.4	57.75	54.14	48.48	57.52	67.69	83.33	67.84	79.14	76.44
tner/roberta-base-tweetner7-continuous	tweetner7	roberta-base	65.47	60.01	50.97	41.68	46.75	61.52	67.98	84.49	66.67	65.15	60.82	58.05	49.85	44.74	56.05	67.08	82.63	67.33	78.1	75.05
tner/twitter-roberta-base-2019-90m-tweetner7-continuous	tweetner7	cardiffnlp/twitter-roberta-base-2019-90m	65.87	61.07	51.66	48.01	48.47	60.42	68.36	84.59	66.01	64.76	60.58	56.19	54.97	44.67	53.17	63.53	83.64	67.88	78.44	75.53
tner/twitter-roberta-base-dec2020-tweetner7-continuous	tweetner7	cardiffnlp/twitter-roberta-base-dec2020	65.51	60.57	53.56	45.3	46.92	61.07	66.28	84.33	66.49	65.29	61.28	59.26	55.59	43.84	54.38	64.14	84.08	67.68	78.03	75.88
tner/bertweet-large-tweetner7-continuous	tweetner7	cardiffnlp/twitter-roberta-base-dec2021vinai/bertweet-large	66.41	61.66	55.07	46.85	48.16	61.44	68.87	84.04	67.18	65.88	61.82	58.38	54.65	46.12	56.39	66.67	83.89	66.67	78.97	76.42
tner/bertweet-base-tweetner7-continuous	tweetner7	vinai/bertweet-base	65.84	61.02	51.85	46.83	49.66	61.17	66.58	84.47	66.59	65.16	61.35	55.76	56.83	46.22	56.32	66.27	82.94	65.13	79.1	76.8
tner/bert-large-tweetner7-continuous	tweetner7	bert-large	63.2	57.67	51.4	39.74	42.55	58.6	63.36	81.27	66.78	62.48	57.87	56.56	43.65	45.51	50.38	60.26	80.62	68.12	76.04	72.46
tner/bert-base-tweetner7-continuous	tweetner7	bert-base	61.8	56.84	47.4	38.22	44.05	57.73	64.42	80.72	65.31	61.41	57.11	54.41	42.41	41.46	51.25	63.49	79.9	66.84	76.53	72.5
tner/roberta-large-tweetner7-2021	tweetner7	roberta-large	64.05	59.11	50.58	43.91	46.6	60.68	63.99	82.68	65.3	63.36	59.15	53.22	49.41	46.61	54.65	63.12	81.33	65.67	77.71	74.36
tner/roberta-base-tweetner7-2021	tweetner7	roberta-base	61.76	57	48.9	38	45.51	57.02	65.06	81.34	63.17	60.5	56.12	49.86	45.33	39.83	52.81	60.95	79.93	64.15	76.92	73.75
tner/twitter-roberta-base-2019-90m-tweetner7-2021	tweetner7	cardiffnlp/twitter-roberta-base-2019-90m	63.23	56.72	46.73	33.12	45.97	57.61	64.42	83.21	65.95	61.91	56.09	48.59	41.1	44.35	49.57	64.16	82.3	62.6	75.69	73.04
tner/twitter-roberta-base-dec2020-tweetner7-2021	tweetner7	cardiffnlp/twitter-roberta-base-dec2020	63.98	58.91	51.04	40.86	46.2	60.22	65.55	82.64	65.88	63.07	58.51	53.26	47.09	40.92	56.46	64.86	82.1	64.89	77.87	75.35
tner/bertweet-large-tweetner7-2021	tweetner7	cardiffnlp/twitter-roberta-base-dec2021vinai/bertweet-large	62.9	58.13	48.87	42.33	44.87	56.4	66.21	81.05	67.16	61.61	56.84	54.24	40.83	43.34	50.3	64.56	81.57	63.05	76.5	74.46
tner/bertweet-base-tweetner7-2021	tweetner7	vinai/bertweet-base	63.09	57.35	45.66	40.99	46.28	59.32	63.34	82.79	63.1	62.06	57.23	49.87	45.83	43.89	52.65	63.58	81.79	63.01	77.88	75.95
tner/bert-large-tweetner7-2021	tweetner7	bert-large	59.75	53.93	44.87	34.17	40.24	55.68	63.95	79.4	59.19	56.63	50.97	49.32	31.58	30.39	50.27	59.76	76.07	59.41	74.98	70.66
tner/bert-base-tweetner7-2021	tweetner7	bert-base	60.67	55.5	46.8	35.35	41.28	56.23	64.78	79.89	64.17	58.45	54.22	48.84	43.05	32.27	50.65	61.54	76.68	66.5	75.72	70.86
tner/roberta-large-tweetner7-2020	tweetner7	roberta-large	64.76	60	52.23	45.89	48.51	60.88	64.43	83.32	64.75	65.67	61.88	56.82	51.85	51.06	58.65	67.06	82.59	65.15	78.36	76.11
tner/roberta-base-tweetner7-2020	tweetner7	roberta-base	64.21	59.11	50.75	44.44	43.9	59.15	65.84	83.92	65.73	64.25	60.23	58.59	48.94	43.84	55.31	65.63	82	67.32	77.89	74.8
tner/twitter-roberta-base-2019-90m-tweetner7-2020	tweetner7	cardiffnlp/twitter-roberta-base-2019-90m	64.28	59.31	48.54	46.89	43.69	59.09	67.01	84	65.98	65.42	61.11	56.28	53.69	43.39	56.23	64.76	84.73	68.72	77.9	76.56
tner/twitter-roberta-base-dec2020-tweetner7-2020	tweetner7	cardiffnlp/twitter-roberta-base-dec2020	62.87	58.26	49.9	44.9	43.68	57.62	64.38	82.29	65.07	64.39	60.31	55.19	51.72	42.91	55.95	65.47	83.98	66.98	76.49	75.65
tner/bertweet-large-tweetner7-2020	tweetner7	cardiffnlp/twitter-roberta-base-dec2021vinai/bertweet-large	64.01	59.47	52.29	46.3	45	59.27	65.53	82.73	65.19	65.93	62.61	59.67	58.92	45.01	54.55	68.09	83.59	68.47	78.26	77.38
tner/bertweet-base-tweetner7-2020	tweetner7	vinai/bertweet-base	64.06	59.44	51.62	45.72	45.87	59.74	64.7	82.71	65.74	66.38	62.41	58.05	54.95	49.9	56.18	67.45	84.75	65.57	77.91	77.73
tner/bert-large-tweetner7-2020	tweetner7	bert-large	61.43	56.14	50.11	39.03	41.8	57.31	61.13	80.6	63.02	62.19	58.15	56.68	43.75	47.24	49.72	62.62	80.03	66.97	75.86	73.79
tner/bert-base-tweetner7-2020	tweetner7	bert-base	60.09	54.67	44.11	37.52	40.28	55.77	61.8	80.52	62.73	60.87	56.49	50.77	44.07	38.35	53.18	63.29	81.06	64.71	75.61	72.42

Model description follows below.

• Model with suffix -all: Model fine-tuned on train_all and validated on validation_2021.
• Model with suffix -continuous: Model fine-tuned on train_2021 continuously after fine-tuning on train_2020 and validated on validation_2021.
• Model with suffix -2021: Model fine-tuned only on train_2021 and validated on validation_2021.
• Model with suffix -2020: Model fine-tuned only on train_2021 and validated on validation_2020.

Sub Models (used in ablation study)

• Model fine-tuned only on train_random and validated on validation_2020.

Model (link)	Data	Language Model	Micro F1 (2021)	Macro F1 (2021)	F1 (2021)/corporation	F1 (2021)/creative_work	F1 (2021)/event	F1 (2021)/group	F1 (2021)/location	F1 (2021)/person	F1 (2021)/product	Micro F1 (2020)	Macro F1 (2020)	F1 (2020)/corporation	F1 (2020)/creative_work	F1 (2020)/event	F1 (2020)/group	F1 (2020)/location	F1 (2020)/person	F1 (2020)/product	Entity-Span F1 (2021)	Entity-Span F1 (2020)
tner/roberta-large-tweetner7-random	tweetner7	roberta-large	66.33	60.96	52.24	45.19	48.95	63.28	66.92	83.84	66.34	64.4	60.09	53.45	50.27	46.68	57.25	65.44	81.79	65.73	79	75.52
tner/roberta-base-tweetner7-random	tweetner7	roberta-base	64.04	59.23	50.73	42.35	45.98	59.73	67.95	82.32	65.58	64.14	59.78	57.58	47.62	42.19	56.48	67.07	82.71	64.84	78.04	74.26
tner/twitter-roberta-base-2019-90m-tweetner7-random	tweetner7	cardiffnlp/twitter-roberta-base-2019-90m	63.29	58.5	50.56	41.68	45.7	59.91	64.8	83.02	63.82	64.29	60.67	56.85	48.88	45.36	55.03	71.75	82.29	64.55	77.36	76.21
tner/twitter-roberta-base-dec2020-tweetner7-random	tweetner7	cardiffnlp/twitter-roberta-base-dec2020	64.72	59.97	49.08	46.42	45.65	61.68	67.5	83.31	66.15	64.69	60.53	55.56	53.85	44.27	56.57	65.05	84.03	64.41	78.29	75.94
tner/bertweet-large-tweetner7-random	tweetner7	cardiffnlp/twitter-roberta-base-dec2021vinai/bertweet-large	64.86	60.49	53.59	45.47	46.19	61.64	66.16	82.79	67.58	66.02	62.72	57.81	58.19	47.64	58.78	68.25	83.36	64.97	78.43	77.2
tner/bertweet-base-tweetner7-random	tweetner7	vinai/bertweet-base	65.55	59.58	49.6	40.06	47.29	62.07	67.98	83.52	66.56	63.89	58.61	54.38	45.05	41.97	55.88	66.03	83.36	63.61	77.8	74.39
tner/bert-large-tweetner7-random	tweetner7	bert-large	62.39	57.54	49.15	39.72	44.79	57.67	67.22	81.17	63.07	61.54	57.09	56.34	42.81	42.69	53.36	61.98	81.04	61.43	76.49	73.29
tner/bert-base-tweetner7-random	tweetner7	bert-base	60.91	55.92	46.51	39.05	41.83	56.14	63.9	80.45	63.54	61.04	56.75	53.94	42.77	39.15	53.07	62.67	80.59	65.08	75.72	72.73

• Model fine-tuned on the self-labeled dataset on extra_{2020,2021} and validated on validation_2020.

Model (link)	Data	Language Model	Micro F1 (2021)	Macro F1 (2021)	F1 (2021)/corporation	F1 (2021)/creative_work	F1 (2021)/event	F1 (2021)/group	F1 (2021)/location	F1 (2021)/person	F1 (2021)/product	Micro F1 (2020)	Macro F1 (2020)	F1 (2020)/corporation	F1 (2020)/creative_work	F1 (2020)/event	F1 (2020)/group	F1 (2020)/location	F1 (2020)/person	F1 (2020)/product	Entity-Span F1 (2021)	Entity-Span F1 (2020)
tner/roberta-large-tweetner7-selflabel2020	tweetner7	roberta-large	64.56	59.63	52.28	46.82	44.47	61.55	64.24	84.02	64.02	65.9	61.85	58.15	51.99	48.05	57.25	66.86	84.16	66.51	78.46	76.71
tner/roberta-large-tweetner7-selflabel2021	tweetner7	roberta-large	64.6	59.45	50.21	45.89	45.18	60.3	66.71	83.46	64.38	64.75	60.65	56.19	50.41	47.31	55.21	67.46	81.9	66.06	78.57	76.63
tner/roberta-large-tweetner7-2020-selflabel2020-all	tweetner7	roberta-large	65.46	60.39	52.56	46.12	45.83	61.7	67.17	84.39	64.95	66.23	62.26	57.5	54.2	46.75	58.32	67.86	83.56	67.61	79.17	77.17
tner/roberta-large-tweetner7-2020-selflabel2021-all	tweetner7	roberta-large	64.52	59.45	50.67	45.38	44.53	60.63	66.19	83.59	65.17	66.05	61.83	58.23	53.44	44.39	59.79	68.09	83.43	65.43	78.5	76.94
tner/roberta-large-tweetner7-selflabel2020-continuous	tweetner7	roberta-large	65.15	60.23	52.53	46.5	46.18	60.87	66.67	83.83	65.03	66.7	62.86	59.35	54.44	48.28	59.44	67.66	83.36	67.45	78.73	77.12
tner/roberta-large-tweetner7-selflabel2021-continuous	tweetner7	roberta-large	64.48	59.41	50.58	45.67	44.4	61.09	66.36	83.63	64.14	65.48	61.42	56.93	51.75	48.72	57.61	67.27	83.29	64.37	78.36	76.5

Model description follows below.

• Model with suffix -self2020: Fine-tuning on the self-annotated data of extra_2020 split of tweetner7.
• Model with suffix -self2021: Fine-tuning on the self-annotated data of extra_2021 split of tweetner7.
• Model with suffix -2020-self2020-all: Fine-tuning on the self-annotated data of extra_2020 split of tweetner7. Combined training dataset of extra_2020 and train_2020.
• Model with suffix -2020-self2021-all: Fine-tuning on the self-annotated data of extra_2021 split of tweetner7. Combined training dataset of extra_2021 and train_2020.
• Model with suffix -2020-self2020-continuous: Fine-tuning on the self-annotated data of extra_2020 split of tweetner7. Fine-tuning on train_2020 and continuing fine-tuning on extra_2020.
• Model with suffix -2020-self2021-continuous: Fine-tuning on the self-annotated data of extra_2021 split of tweetner7. Fine-tuning on train_2020 and continuing fine-tuning on extra_2020.

Reproduce Experimental Result

To reproduce the experimental result on our AACL paper, please see the repository https://github.com/asahi417/tner/tree/master/examples/tweetner7_paper.

Citation Information

@inproceedings{ushio-etal-2022-tweet,
    title = "{N}amed {E}ntity {R}ecognition in {T}witter: {A} {D}ataset and {A}nalysis on {S}hort-{T}erm {T}emporal {S}hifts",
    author = "Ushio, Asahi  and
        Neves, Leonardo  and
        Silva, Vitor  and
        Barbieri, Francesco. and
        Camacho-Collados, Jose",
    booktitle = "The 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing",
    month = nov,
    year = "2022",
    address = "Online",
    publisher = "Association for Computational Linguistics",
}