Update dataset card
Browse files
README.md
CHANGED
|
@@ -26,7 +26,9 @@ configs:
|
|
| 26 |
path: data/dev_test-*
|
| 27 |
- split: test
|
| 28 |
path: data/test-*
|
|
|
|
| 29 |
dataset_info:
|
|
|
|
| 30 |
features:
|
| 31 |
- name: id
|
| 32 |
dtype: string
|
|
@@ -40,19 +42,13 @@ dataset_info:
|
|
| 40 |
sequence: string
|
| 41 |
splits:
|
| 42 |
- name: train
|
| 43 |
-
num_bytes: 274812860.5
|
| 44 |
num_examples: 3500
|
| 45 |
- name: dev
|
| 46 |
-
num_bytes: 40130069.0
|
| 47 |
num_examples: 500
|
| 48 |
- name: dev_test
|
| 49 |
-
num_bytes: 38915856.0
|
| 50 |
num_examples: 500
|
| 51 |
- name: test
|
| 52 |
-
num_bytes: 38669448.0
|
| 53 |
num_examples: 500
|
| 54 |
-
download_size: 386298290
|
| 55 |
-
dataset_size: 392528233.5
|
| 56 |
---
|
| 57 |
|
| 58 |
# ArGuard – Track A: Arabic Hateful Memes
|
|
@@ -81,30 +77,33 @@ Given a meme (image + Arabic text):
|
|
| 81 |
|
| 82 |
## Splits
|
| 83 |
|
| 84 |
-
| Split | Records | Labels | Source
|
| 85 |
-
|--------------|---------|--------------|----------------------------------------|---------------------------|
|
| 86 |
-
| `train` | 3,500 | full | single-annotated bulk
|
| 87 |
-
| `dev` | 500 | full | single-annotated bulk
|
| 88 |
-
| `dev_test` | 500 | **dropped** | single-annotated test sample
|
| 89 |
-
| `test` | 500 |
|
| 90 |
-
|
| 91 |
-
- `dev_test`
|
| 92 |
-
|
| 93 |
-
|
| 94 |
-
- `test` is the **
|
| 95 |
-
|
| 96 |
-
|
| 97 |
-
|
|
|
|
| 98 |
|
| 99 |
### Binary label distribution
|
| 100 |
|
| 101 |
| Split | Hateful | Not Hateful | % Hateful |
|
| 102 |
|------------|--------:|------------:|----------:|
|
| 103 |
| train | 1,324 | 2,176 | 37.8% |
|
| 104 |
-
| dev |
|
| 105 |
-
| dev_test |
|
| 106 |
-
|
|
| 107 |
-
|
|
|
|
|
|
|
| 108 |
|
| 109 |
### Fine-grained sub-types (Subtask A2)
|
| 110 |
|
|
@@ -134,8 +133,8 @@ than the meme counts.
|
|
| 134 |
"id": "f9a8…b1.jpg", # str – original image filename, unique
|
| 135 |
"image": <PIL.Image.Image>, # embedded bytes, decoded on access
|
| 136 |
"text": "…", # str – OCR-extracted Arabic meme text
|
| 137 |
-
"label": "Hateful" | "Not Hateful" | None, # None on dev_test
|
| 138 |
-
"fine_grained_label": [...], # list[str] – empty on dev_test
|
| 139 |
}
|
| 140 |
```
|
| 141 |
|
|
@@ -151,14 +150,20 @@ train_ex = ds["train"][0]
|
|
| 151 |
train_ex["image"].show()
|
| 152 |
print(train_ex["text"], train_ex["label"], train_ex["fine_grained_label"])
|
| 153 |
|
| 154 |
-
# dev_test
|
| 155 |
-
print(ds["
|
| 156 |
```
|
| 157 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 158 |
## Shared-task resources
|
| 159 |
|
| 160 |
-
- **Website:** https://araieval.
|
| 161 |
- **Starter kit / baselines / scorers (GitHub):** https://github.com/araieval/ArGuard-2026-tasks
|
|
|
|
|
|
|
| 162 |
- **Contact organisers:** arguard2026-organizers@googlegroups.com
|
| 163 |
|
| 164 |
## Annotation
|
|
@@ -191,29 +196,11 @@ generate harmful content.
|
|
| 191 |
|
| 192 |
## Citation
|
| 193 |
|
| 194 |
-
|
| 195 |
-
|
| 196 |
-
title = {{ArGuard Shared Task}: Harmful Content Detection in
|
| 197 |
-
{A}rabic {M}emes and {LLM} {P}rompts},
|
| 198 |
-
author = {Alam, Firoj and
|
| 199 |
-
Biswas, Md. Rafiul and
|
| 200 |
-
Kmainasi, Mohamed Bayan and
|
| 201 |
-
Shahroor, Ali Ezzat and
|
| 202 |
-
Mubarak, Hamdy and
|
| 203 |
-
Mikros, Georgios and
|
| 204 |
-
Hasnat, Abul and
|
| 205 |
-
Zaghouani, Wajdi},
|
| 206 |
-
booktitle = {Proceedings of the Fourth Arabic Natural Language
|
| 207 |
-
Processing Conference: Shared Tasks},
|
| 208 |
-
month = oct,
|
| 209 |
-
year = {2026},
|
| 210 |
-
address = {Budapest, Hungary},
|
| 211 |
-
publisher = {Association for Computational Linguistics}
|
| 212 |
-
}
|
| 213 |
-
```
|
| 214 |
|
| 215 |
## Contact
|
| 216 |
|
| 217 |
- **Email:** arguard2026-organizers@googlegroups.com
|
| 218 |
-
- **Website:** https://araieval.
|
| 219 |
- **GitHub:** https://github.com/araieval/ArGuard-2026-tasks
|
|
|
|
| 26 |
path: data/dev_test-*
|
| 27 |
- split: test
|
| 28 |
path: data/test-*
|
| 29 |
+
default: true
|
| 30 |
dataset_info:
|
| 31 |
+
config_name: default
|
| 32 |
features:
|
| 33 |
- name: id
|
| 34 |
dtype: string
|
|
|
|
| 42 |
sequence: string
|
| 43 |
splits:
|
| 44 |
- name: train
|
|
|
|
| 45 |
num_examples: 3500
|
| 46 |
- name: dev
|
|
|
|
| 47 |
num_examples: 500
|
| 48 |
- name: dev_test
|
|
|
|
| 49 |
num_examples: 500
|
| 50 |
- name: test
|
|
|
|
| 51 |
num_examples: 500
|
|
|
|
|
|
|
| 52 |
---
|
| 53 |
|
| 54 |
# ArGuard – Track A: Arabic Hateful Memes
|
|
|
|
| 77 |
|
| 78 |
## Splits
|
| 79 |
|
| 80 |
+
| Split | Records | Labels | Source | Released |
|
| 81 |
+
|--------------|---------|--------------|-----------------------------------------|---------------------------------|
|
| 82 |
+
| `train` | 3,500 | full | single-annotated bulk | development phase |
|
| 83 |
+
| `dev` | 500 | full | single-annotated bulk | development phase |
|
| 84 |
+
| `dev_test` | 500 | **dropped** | single-annotated test sample | development phase (leaderboard) |
|
| 85 |
+
| `test` | 500 | **dropped** | **triple-annotated gold** (calibration) | final-evaluation phase (leaderboard) |
|
| 86 |
+
|
| 87 |
+
- `dev_test` was the **leaderboard set** for the development phase, which
|
| 88 |
+
is now closed. Labels are stripped (`label = null`,
|
| 89 |
+
`fine_grained_label = []`).
|
| 90 |
+
- `test` is the **blind test set** for the final-evaluation phase and is
|
| 91 |
+
the file you submit predictions for. It is released the same way as
|
| 92 |
+
`dev_test` — images and OCR text only, labels stripped. All 500 records
|
| 93 |
+
are triple-annotated with majority voting; the gold labels are held in
|
| 94 |
+
the CodaBench reference data and are never published here.
|
| 95 |
|
| 96 |
### Binary label distribution
|
| 97 |
|
| 98 |
| Split | Hateful | Not Hateful | % Hateful |
|
| 99 |
|------------|--------:|------------:|----------:|
|
| 100 |
| train | 1,324 | 2,176 | 37.8% |
|
| 101 |
+
| dev | 189 | 311 | 37.8% |
|
| 102 |
+
| dev_test | 189 | 311 | 37.8% |
|
| 103 |
+
| **Total** | **1,702** | **2,798** | 37.8% |
|
| 104 |
+
|
| 105 |
+
The label distribution of the blind `test` split is withheld during the
|
| 106 |
+
final-evaluation phase.
|
| 107 |
|
| 108 |
### Fine-grained sub-types (Subtask A2)
|
| 109 |
|
|
|
|
| 133 |
"id": "f9a8…b1.jpg", # str – original image filename, unique
|
| 134 |
"image": <PIL.Image.Image>, # embedded bytes, decoded on access
|
| 135 |
"text": "…", # str – OCR-extracted Arabic meme text
|
| 136 |
+
"label": "Hateful" | "Not Hateful" | None, # None on dev_test / test
|
| 137 |
+
"fine_grained_label": [...], # list[str] – empty on dev_test / test
|
| 138 |
}
|
| 139 |
```
|
| 140 |
|
|
|
|
| 150 |
train_ex["image"].show()
|
| 151 |
print(train_ex["text"], train_ex["label"], train_ex["fine_grained_label"])
|
| 152 |
|
| 153 |
+
# dev_test and test are unlabelled — used only to produce submissions
|
| 154 |
+
print(ds["test"][0]["label"]) # -> None
|
| 155 |
```
|
| 156 |
|
| 157 |
+
The blind `test` split is the submission target for the final-evaluation
|
| 158 |
+
phase. Train on `train` + `dev`, predict on `test`, and upload to
|
| 159 |
+
CodaBench (links below).
|
| 160 |
+
|
| 161 |
## Shared-task resources
|
| 162 |
|
| 163 |
+
- **Website:** https://araieval.github.io/ArGuard2026/
|
| 164 |
- **Starter kit / baselines / scorers (GitHub):** https://github.com/araieval/ArGuard-2026-tasks
|
| 165 |
+
- **Submission — Subtask A1:** https://www.codabench.org/competitions/16909/
|
| 166 |
+
- **Submission — Subtask A2:** https://www.codabench.org/competitions/16910/
|
| 167 |
- **Contact organisers:** arguard2026-organizers@googlegroups.com
|
| 168 |
|
| 169 |
## Annotation
|
|
|
|
| 196 |
|
| 197 |
## Citation
|
| 198 |
|
| 199 |
+
A citation will be provided when the shared-task overview paper is
|
| 200 |
+
released. Until then, please cite this repository URL.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 201 |
|
| 202 |
## Contact
|
| 203 |
|
| 204 |
- **Email:** arguard2026-organizers@googlegroups.com
|
| 205 |
+
- **Website:** https://araieval.github.io/ArGuard2026/
|
| 206 |
- **GitHub:** https://github.com/araieval/ArGuard-2026-tasks
|