metadata

license: cc-by-nc-sa-4.0
task_categories:
  - image-classification
  - text-classification
  - visual-question-answering
language:
  - ar
pretty_name: ArMeme
size_categories:
  - 1K<n<10K
dataset_info:
  features:
    - name: id
      dtype: string
    - name: text
      dtype: string
    - name: image
      dtype: image
    - name: img_path
      dtype: string
    - name: class_label
      dtype:
        class_label:
          names:
            '0': not_propaganda
            '1': propaganda
            '2': not-meme
            '3': other
  splits:
    - name: train
      num_bytes: 288878900.171
      num_examples: 4007
    - name: dev
      num_bytes: 45908447
      num_examples: 584
    - name: test
      num_bytes: 81787436.176
      num_examples: 1134
  download_size: 423396230
  dataset_size: 416574783.347

ArMeme Dataset

Overview

ArMeme is the first multimodal Arabic memes dataset that includes both text and images, collected from various social media platforms. It serves as the first resource dedicated to Arabic multimodal research. While the dataset has been annotated to identify propaganda in memes, it is versatile and can be utilized for a wide range of other research purposes, including sentiment analysis, hate speech detection, cultural studies, meme generation, and cross-lingual transfer learning. The dataset opens new avenues for exploring the intersection of language, culture, and visual communication.

Dataset Structure

The dataset is divided into three splits:

Train: The training set
Dev: The development/validation set
Test: The test set

Each entry in the dataset includes:

id: id corresponds to the entry
text: The textual content associated with the image.
image: The corresponding image.
img_path: The file path to the image.

How to Use

You can load the dataset using the datasets library from Hugging Face:

from datasets import load_dataset

dataset = load_dataset("QCRI/ArMeme")

# Specify the directory where you want to save the dataset
output_dir="./ArMeme/"

# Save the dataset to the specified directory. This will save all splits to the output directory.
dataset.save_to_disk(output_dir)

# If you want to get the raw images from HF dataset format

from PIL import Image
import os
import json

# Directory to save the images
output_dir="./ArMeme/"
os.makedirs(output_dir, exist_ok=True)

# Iterate over the dataset and save each image
for split in ['train','dev','test']:     
    jsonl_path = os.path.join(output_dir, f"arabic_memes_categorization_{split}.jsonl")
    with open(jsonl_path, 'w', encoding='utf-8') as f:    
        for idx, item in enumerate(dataset[split]):
            # Access the image directly as it's already a PIL.Image object
            image = item['image']
            image_path = os.path.join(output_dir, item['img_path'])
            # Ensure the directory exists
            os.makedirs(os.path.dirname(image_path), exist_ok=True)
            image.save(image_path)
            del item['image']
            f.write(json.dumps(item, ensure_ascii=False) + '\n')

Language: Arabic

Modality: Multimodal (text + image)

Number of Samples: ~6000

License

This dataset is licensed under the CC-By-NC-SA-4.0 license.

Citation

Please find the paper on ArXiv and use the bib info below to cite the paper.

@inproceedings{alam2024armeme,
  title={{ArMeme}: Propagandistic Content in Arabic Memes},
  author={Alam, Firoj and Hasnat, Abul and Ahmed, Fatema and Hasan, Md Arid and Hasanain, Maram},
  booktitle={Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year={2024},
  address={Miami, Florida},
  month={November 12--16},
  publisher={Association for Computational Linguistics},
}