asahi417 commited on
Commit
dbb8f50
1 Parent(s): 9c2a531

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +196 -0
README.md ADDED
@@ -0,0 +1,196 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language:
3
+ - en
4
+ - ar
5
+ - fr
6
+ - de
7
+ - hi
8
+ - it
9
+ - pt
10
+ - es
11
+ multilinguality:
12
+ - multilingual
13
+ size_categories:
14
+ - 10K<n<100K
15
+ source_datasets:
16
+ - extended|other-tweet-datasets
17
+ task_categories:
18
+ - text-classification
19
+ task_ids:
20
+ - sentiment-classification
21
+ paperswithcode_id: tweet_sentiment_multilingual
22
+ pretty_name: Tweet Sentiment Multilingual
23
+ train-eval-index:
24
+ - config: sentiment
25
+ task: text-classification
26
+ task_id: multi_class_classification
27
+ splits:
28
+ train_split: train
29
+ eval_split: test
30
+ col_mapping:
31
+ text: text
32
+ label: target
33
+ metrics:
34
+ - type: accuracy
35
+ name: Accuracy
36
+ - type: f1
37
+ name: F1 macro
38
+ args:
39
+ average: macro
40
+ - type: f1
41
+ name: F1 micro
42
+ args:
43
+ average: micro
44
+ - type: f1
45
+ name: F1 weighted
46
+ args:
47
+ average: weighted
48
+ - type: precision
49
+ name: Precision macro
50
+ args:
51
+ average: macro
52
+ - type: precision
53
+ name: Precision micro
54
+ args:
55
+ average: micro
56
+ - type: precision
57
+ name: Precision weighted
58
+ args:
59
+ average: weighted
60
+ - type: recall
61
+ name: Recall macro
62
+ args:
63
+ average: macro
64
+ - type: recall
65
+ name: Recall micro
66
+ args:
67
+ average: micro
68
+ - type: recall
69
+ name: Recall weighted
70
+ args:
71
+ average: weighted
72
+ configs:
73
+ - arabic
74
+ - english
75
+ - french
76
+ - german
77
+ - hindi
78
+ - italian
79
+ - portuguese
80
+ - spanish
81
+ dataset_info:
82
+ - config_name: sentiment
83
+ features:
84
+ - name: text
85
+ dtype: string
86
+ - name: label
87
+ dtype:
88
+ class_label:
89
+ names:
90
+ 0: negative
91
+ 1: neutral
92
+ 2: positive
93
+ ---
94
+
95
+ # Dataset Card for cardiffnlp/tweet_sentiment_multilingual
96
+
97
+ ## Dataset Description
98
+
99
+ - **Homepage:** [https://github.com/cardiffnlp/xlm-t](https://github.com/cardiffnlp/xlm-t)
100
+ - **Repository:** - **Homepage:** [https://github.com/cardiffnlp/xlm-t](https://github.com/cardiffnlp/xlm-t)
101
+ - **Paper:** [https://aclanthology.org/2022.lrec-1.27/](https://aclanthology.org/2022.lrec-1.27/)
102
+ - **Point of Contact:** [Asahi Ushio](https://asahiushio.com/)
103
+
104
+ ### Dataset Summary
105
+
106
+ Tweet Sentiment Multilingual consists of sentiment analysis dataset on Twitter in 8 different lagnuages.
107
+
108
+ - arabic
109
+ - english
110
+ - french
111
+ - german
112
+ - hindi
113
+ - italian
114
+ - portuguese
115
+ - spanish
116
+
117
+ ### Supported Tasks and Leaderboards
118
+
119
+ - `text_classification`: The dataset can be trained using a SentenceClassification model from HuggingFace transformers.
120
+
121
+ ## Dataset Structure
122
+
123
+ ### Data Instances
124
+
125
+ An instance from `sentiment` config:
126
+
127
+ ```
128
+ {'label': 2, 'text': '"QT @user In the original draft of the 7th book, Remus Lupin survived the Battle of Hogwarts. #HappyBirthdayRemusLupin"'}
129
+ ```
130
+
131
+ ### Data Fields
132
+
133
+ For `sentiment` config:
134
+
135
+ - `text`: a `string` feature containing the tweet.
136
+
137
+ - `label`: an `int` classification label with the following mapping:
138
+
139
+ `0`: negative
140
+
141
+ `1`: neutral
142
+
143
+ `2`: positive
144
+
145
+
146
+ ### Data Splits
147
+
148
+
149
+ - arabic
150
+ - english
151
+ - french
152
+ - german
153
+ - hindi
154
+ - italian
155
+ - portuguese
156
+ - spanish
157
+
158
+ | name | train | validation | test |
159
+ | --------------- | ----- | ---------- | ----- |
160
+ | arabic | 1838 | 323 | 869 |
161
+ | english | 1838 | 323 | 869 |
162
+ | french | 1838 | 323 | 869 |
163
+ | german | 1838 | 323 | 869 |
164
+ | hindi | 1838 | 323 | 869 |
165
+ | italian | 1838 | 323 | 869 |
166
+ | portuguese | 1838 | 323 | 869 |
167
+ | spanish | 1838 | 323 | 869 |
168
+
169
+ ### Dataset Curators
170
+
171
+ Francesco Barbieri, Jose Camacho-Collados, Luis Espiinosa-Anke and Leonardo Neves through Cardiff NLP.
172
+
173
+ ### Licensing Information
174
+
175
+ [Creative Commons Attribution 3.0 Unported License](https://groups.google.com/g/semevaltweet/c/k5DDcvVb_Vo/m/zEOdECFyBQAJ), and all of the datasets require complying with Twitter [Terms Of Service](https://twitter.com/tos) and Twitter API [Terms Of Service](https://developer.twitter.com/en/developer-terms/agreement-and-policy)
176
+
177
+
178
+ ### Citation Information
179
+
180
+ ```
181
+ @inproceedings{barbieri-etal-2022-xlm,
182
+ title = "{XLM}-{T}: Multilingual Language Models in {T}witter for Sentiment Analysis and Beyond",
183
+ author = "Barbieri, Francesco and
184
+ Espinosa Anke, Luis and
185
+ Camacho-Collados, Jose",
186
+ booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
187
+ month = jun,
188
+ year = "2022",
189
+ address = "Marseille, France",
190
+ publisher = "European Language Resources Association",
191
+ url = "https://aclanthology.org/2022.lrec-1.27",
192
+ pages = "258--266",
193
+ abstract = "Language models are ubiquitous in current NLP, and their multilingual capacity has recently attracted considerable attention. However, current analyses have almost exclusively focused on (multilingual variants of) standard benchmarks, and have relied on clean pre-training and task-specific corpora as multilingual signals. In this paper, we introduce XLM-T, a model to train and evaluate multilingual language models in Twitter. In this paper we provide: (1) a new strong multilingual baseline consisting of an XLM-R (Conneau et al. 2020) model pre-trained on millions of tweets in over thirty languages, alongside starter code to subsequently fine-tune on a target task; and (2) a set of unified sentiment analysis Twitter datasets in eight different languages and a XLM-T model trained on this dataset.",
194
+ }
195
+ ```
196
+