Datasets:
QCRI
/

Modalities:
Text
Formats:
json
Languages:
Arabic
ArXiv:
Libraries:
Datasets
pandas
License:
MohamedBayan commited on
Commit
3b57c69
1 Parent(s): 4ba2975

Adding readme

Browse files
Files changed (1) hide show
  1. README.md +384 -0
README.md ADDED
@@ -0,0 +1,384 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-sa-4.0
3
+ task_categories:
4
+ - text-classification
5
+ language:
6
+ - ar
7
+ tags:
8
+ - Social Media
9
+ - News Media
10
+ - Sentiment
11
+ - Stance
12
+ - Emotion
13
+ pretty_name: 'LlamaLens: Specialized Multilingual LLM for Analyzing News and Social Media Content -- English'
14
+ size_categories:
15
+ - 10K<n<100K
16
+ dataset_info:
17
+ - config_name: QProp
18
+ splits:
19
+ - name: train
20
+ num_examples: 35986
21
+ - name: dev
22
+ num_examples: 5125
23
+ - name: test
24
+ num_examples: 10159
25
+ - config_name: Cyberbullying
26
+ splits:
27
+ - name: train
28
+ num_examples: 32551
29
+ - name: dev
30
+ num_examples: 4751
31
+ - name: test
32
+ num_examples: 9473
33
+ - config_name: clef2024-checkthat-lab
34
+ splits:
35
+ - name: train
36
+ num_examples: 825
37
+ - name: dev
38
+ num_examples: 219
39
+ - name: test
40
+ num_examples: 484
41
+ - config_name: SemEval23T3-subtask1
42
+ splits:
43
+ - name: train
44
+ num_examples: 302
45
+ - name: dev
46
+ num_examples: 130
47
+ - name: test
48
+ num_examples: 83
49
+ - config_name: offensive_language_dataset
50
+ splits:
51
+ - name: train
52
+ num_examples: 29216
53
+ - name: dev
54
+ num_examples: 3653
55
+ - name: test
56
+ num_examples: 3653
57
+ - config_name: xlsum
58
+ splits:
59
+ - name: train
60
+ num_examples: 306493
61
+ - name: dev
62
+ num_examples: 11535
63
+ - name: test
64
+ num_examples: 11535
65
+ - config_name: claim-detection
66
+ splits:
67
+ - name: train
68
+ num_examples: 23224
69
+ - name: dev
70
+ num_examples: 5815
71
+ - name: test
72
+ num_examples: 7267
73
+ - config_name: emotion
74
+ splits:
75
+ - name: train
76
+ num_examples: 280551
77
+ - name: dev
78
+ num_examples: 41429
79
+ - name: test
80
+ num_examples: 82454
81
+ - config_name: Politifact
82
+ splits:
83
+ - name: train
84
+ num_examples: 14799
85
+ - name: dev
86
+ num_examples: 2116
87
+ - name: test
88
+ num_examples: 4230
89
+ - config_name: News_dataset
90
+ splits:
91
+ - name: train
92
+ num_examples: 28147
93
+ - name: dev
94
+ num_examples: 4376
95
+ - name: test
96
+ num_examples: 8616
97
+ - config_name: hate-offensive-speech
98
+ splits:
99
+ - name: train
100
+ num_examples: 48944
101
+ - name: dev
102
+ num_examples: 2802
103
+ - name: test
104
+ num_examples: 2799
105
+ - config_name: CNN_News_Articles_2011-2022
106
+ splits:
107
+ - name: train
108
+ num_examples: 32193
109
+ - name: dev
110
+ num_examples: 9663
111
+ - name: test
112
+ num_examples: 5682
113
+ - config_name: CT24_checkworthy
114
+ splits:
115
+ - name: train
116
+ num_examples: 22403
117
+ - name: dev
118
+ num_examples: 318
119
+ - name: test
120
+ num_examples: 1031
121
+ - config_name: News_Category_Dataset
122
+ splits:
123
+ - name: train
124
+ num_examples: 145748
125
+ - name: dev
126
+ num_examples: 20899
127
+ - name: test
128
+ num_examples: 41740
129
+ - config_name: NewsMTSC-dataset
130
+ splits:
131
+ - name: train
132
+ num_examples: 7739
133
+ - name: dev
134
+ num_examples: 320
135
+ - name: test
136
+ num_examples: 747
137
+ - config_name: Offensive_Hateful_Dataset_New
138
+ splits:
139
+ - name: train
140
+ num_examples: 42000
141
+ - name: dev
142
+ num_examples: 5254
143
+ - name: test
144
+ num_examples: 5252
145
+ - config_name: News-Headlines-Dataset-For-Sarcasm-Detection
146
+ splits:
147
+ - name: train
148
+ num_examples: 19965
149
+ - name: dev
150
+ num_examples: 2858
151
+ - name: test
152
+ num_examples: 5719
153
+ configs:
154
+ - config_name: QProp
155
+ data_files:
156
+ - split: test
157
+ path: QProp/test.json
158
+ - split: dev
159
+ path: QProp/dev.json
160
+ - split: train
161
+ path: QProp/train.json
162
+ - config_name: Cyberbullying
163
+ data_files:
164
+ - split: test
165
+ path: Cyberbullying/test.json
166
+ - split: dev
167
+ path: Cyberbullying/dev.json
168
+ - split: train
169
+ path: Cyberbullying/train.json
170
+ - config_name: clef2024-checkthat-lab
171
+ data_files:
172
+ - split: test
173
+ path: clef2024-checkthat-lab/test.json
174
+ - split: dev
175
+ path: clef2024-checkthat-lab/dev.json
176
+ - split: train
177
+ path: clef2024-checkthat-lab/train.json
178
+ - config_name: SemEval23T3-subtask1
179
+ data_files:
180
+ - split: test
181
+ path: SemEval23T3-subtask1/test.json
182
+ - split: dev
183
+ path: SemEval23T3-subtask1/dev.json
184
+ - split: train
185
+ path: SemEval23T3-subtask1/train.json
186
+ - config_name: offensive_language_dataset
187
+ data_files:
188
+ - split: test
189
+ path: offensive_language_dataset/test.json
190
+ - split: dev
191
+ path: offensive_language_dataset/dev.json
192
+ - split: train
193
+ path: offensive_language_dataset/train.json
194
+ - config_name: xlsum
195
+ data_files:
196
+ - split: test
197
+ path: xlsum/test.json
198
+ - split: dev
199
+ path: xlsum/dev.json
200
+ - split: train
201
+ path: xlsum/train.json
202
+ - config_name: claim-detection
203
+ data_files:
204
+ - split: test
205
+ path: claim-detection/test.json
206
+ - split: dev
207
+ path: claim-detection/dev.json
208
+ - split: train
209
+ path: claim-detection/train.json
210
+ - config_name: emotion
211
+ data_files:
212
+ - split: test
213
+ path: emotion/test.json
214
+ - split: dev
215
+ path: emotion/dev.json
216
+ - split: train
217
+ path: emotion/train.json
218
+ - config_name: Politifact
219
+ data_files:
220
+ - split: test
221
+ path: Politifact/test.json
222
+ - split: dev
223
+ path: Politifact/dev.json
224
+ - split: train
225
+ path: Politifact/train.json
226
+ - config_name: News_dataset
227
+ data_files:
228
+ - split: test
229
+ path: News_dataset/test.json
230
+ - split: dev
231
+ path: News_dataset/dev.json
232
+ - split: train
233
+ path: News_dataset/train.json
234
+ - config_name: hate-offensive-speech
235
+ data_files:
236
+ - split: test
237
+ path: hate-offensive-speech/test.json
238
+ - split: dev
239
+ path: hate-offensive-speech/dev.json
240
+ - split: train
241
+ path: hate-offensive-speech/train.json
242
+ - config_name: CNN_News_Articles_2011-2022
243
+ data_files:
244
+ - split: test
245
+ path: CNN_News_Articles_2011-2022/test.json
246
+ - split: dev
247
+ path: CNN_News_Articles_2011-2022/dev.json
248
+ - split: train
249
+ path: CNN_News_Articles_2011-2022/train.json
250
+ - config_name: CT24_checkworthy
251
+ data_files:
252
+ - split: test
253
+ path: CT24_checkworthy/test.json
254
+ - split: dev
255
+ path: CT24_checkworthy/dev.json
256
+ - split: train
257
+ path: CT24_checkworthy/train.json
258
+ - config_name: News_Category_Dataset
259
+ data_files:
260
+ - split: test
261
+ path: News_Category_Dataset/test.json
262
+ - split: dev
263
+ path: News_Category_Dataset/dev.json
264
+ - split: train
265
+ path: News_Category_Dataset/train.json
266
+ - config_name: NewsMTSC-dataset
267
+ data_files:
268
+ - split: test
269
+ path: NewsMTSC-dataset/test.json
270
+ - split: dev
271
+ path: NewsMTSC-dataset/dev.json
272
+ - split: train
273
+ path: NewsMTSC-dataset/train.json
274
+ - config_name: Offensive_Hateful_Dataset_New
275
+ data_files:
276
+ - split: test
277
+ path: Offensive_Hateful_Dataset_New/test.json
278
+ - split: dev
279
+ path: Offensive_Hateful_Dataset_New/dev.json
280
+ - split: train
281
+ path: Offensive_Hateful_Dataset_New/train.json
282
+ - config_name: News-Headlines-Dataset-For-Sarcasm-Detection
283
+ data_files:
284
+ - split: test
285
+ path: News-Headlines-Dataset-For-Sarcasm-Detection/test.json
286
+ - split: dev
287
+ path: News-Headlines-Dataset-For-Sarcasm-Detection/dev.json
288
+ - split: train
289
+ path: News-Headlines-Dataset-For-Sarcasm-Detection/train.json
290
+ ---
291
+
292
+ # LlamaLens: Specialized Multilingual LLM Dataset
293
+
294
+ ## Overview
295
+ LlamaLens is a specialized multilingual LLM designed for analyzing news and social media content. It focuses on 19 NLP tasks, leveraging 52 datasets across Arabic, English, and Hindi.
296
+
297
+
298
+ <p align="center"> <img src="./capablities_tasks_datasets.png" style="width: 40%;" id="title-icon"> </p>
299
+
300
+ ## LlamaLens
301
+ This repo includes scripts needed to run our full pipeline, including data preprocessing and sampling, instruction dataset creation, model fine-tuning, inference and evaluation.
302
+
303
+ ### Features
304
+ - Multilingual support (Arabic, English, Hindi)
305
+ - 19 NLP tasks with 52 datasets
306
+ - Optimized for news and social media content analysis
307
+
308
+ ## 📂 Dataset Overview
309
+
310
+ ### English Datasets
311
+
312
+ | **Task** | **Dataset** | **# Labels** | **# Train** | **# Test** | **# Dev** |
313
+ |---------------------------|------------------------------|--------------|-------------|------------|-----------|
314
+ | Checkworthiness | CT24_T1 | 2 | 22,403 | 1,031 | 318 |
315
+ | Claim | claim-detection | 2 | 23,224 | 7,267 | 5,815 |
316
+ | Cyberbullying | Cyberbullying | 6 | 32,551 | 9,473 | 4,751 |
317
+ | Emotion | emotion | 6 | 280,551 | 82,454 | 41,429 |
318
+ | Factuality | News_dataset | 2 | 28,147 | 8,616 | 4,376 |
319
+ | Factuality | Politifact | 6 | 14,799 | 4,230 | 2,116 |
320
+ | News Genre Categorization | CNN_News_Articles_2011-2022 | 6 | 32,193 | 5,682 | 9,663 |
321
+ | News Genre Categorization | News_Category_Dataset | 42 | 145,748 | 41,740 | 20,899 |
322
+ | News Genre Categorization | SemEval23T3-subtask1 | 3 | 302 | 83 | 130 |
323
+ | Summarization | xlsum | -- | 306,493 | 11,535 | 11,535 |
324
+ | Offensive Language | Offensive_Hateful_Dataset_New | 2 | 42,000 | 5,252 | 5,254 |
325
+ | Offensive Language | offensive_language_dataset | 2 | 29,216 | 3,653 | 3,653 |
326
+ | Offensive/Hate-Speech | hate-offensive-speech | 3 | 48,944 | 2,799 | 2,802 |
327
+ | Propaganda | QProp | 2 | 35,986 | 10,159 | 5,125 |
328
+ | Sarcasm | News-Headlines-Dataset-For-Sarcasm-Detection | 2 | 19,965 | 5,719 | 2,858 |
329
+ | Sentiment | NewsMTSC-dataset | 3 | 7,739 | 747 | 320 |
330
+ | Subjectivity | clef2024-checkthat-lab | 2 | 825 | 484 | 219 |
331
+
332
+
333
+ ## File Format
334
+
335
+ Each JSONL file in the dataset follows a structured format with the following fields:
336
+
337
+ - `id`: Unique identifier for each data entry.
338
+ - `original_id`: Identifier from the original dataset, if available.
339
+ - `input`: The original text that needs to be analyzed.
340
+ - `output`: The label assigned to the text after analysis.
341
+ - `dataset`: Name of the dataset the entry belongs.
342
+ - `task`: The specific task type.
343
+ - `lang`: The language of the input text.
344
+ - `instructions`: A brief set of instructions describing how the text should be labeled.
345
+ - `text`: A formatted structure including instructions and response for the task in a conversation format between the system, user, and assistant, showing the decision process.
346
+
347
+
348
+ **Example entry in JSONL file:**
349
+
350
+ ```
351
+ {
352
+ "id": "3fe3eb6a-843e-4a03-b38c-8333c052f4c4",
353
+ "original_id": "nan",
354
+ "input": "You know, I saw a movie - \"Crocodile Dundee.\"",
355
+ "output": "not_checkworthy",
356
+ "dataset": "CT24_checkworthy",
357
+ "task": "Checkworthiness",
358
+ "lang": "en",
359
+ "instructions": "Analyze the given text and label it as 'checkworthy' if it includes a factual statement that is significant or relevant to verify, or 'not_checkworthy' if it's not worth checking. Return only the label without any explanation, justification or additional text.",
360
+ "text": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>You are a social media expert providing accurate analysis and insights.<|eot_id|><|start_header_id|>user<|end_header_id|>Analyze the given text and label it as 'checkworthy' if it includes a factual statement that is significant or relevant to verify, or 'not_checkworthy' if it's not worth checking. Return only the label without any explanation, justification or additional text.\ninput: You know, I saw a movie - \"Crocodile Dundee.\"\nlabel: <|eot_id|><|start_header_id|>assistant<|end_header_id|>not_checkworthy<|eot_id|><|end_of_text|>"
361
+ }
362
+
363
+ ```
364
+
365
+
366
+ ## 📢 Citation
367
+
368
+ If you use this dataset, please cite our [paper](https://arxiv.org/pdf/2410.15308):
369
+
370
+ ```
371
+ @article{kmainasi2024llamalensspecializedmultilingualllm,
372
+ title={LlamaLens: Specialized Multilingual LLM for Analyzing News and Social Media Content},
373
+ author={Mohamed Bayan Kmainasi and Ali Ezzat Shahroor and Maram Hasanain and Sahinur Rahman Laskar and Naeemul Hassan and Firoj Alam},
374
+ year={2024},
375
+ journal={arXiv preprint arXiv:2410.15308},
376
+ volume={},
377
+ number={},
378
+ pages={},
379
+ url={https://arxiv.org/abs/2410.15308},
380
+ eprint={2410.15308},
381
+ archivePrefix={arXiv},
382
+ primaryClass={cs.CL}
383
+ }
384
+ ```