QCRI/ThatiAR

---
license: cc-by-nc-sa-4.0
task_categories:
- text-classification
language:
- ar
tags:
- subjectivity
- sentiment
pretty_name: 'ThatiAR: Subjectivity Detection in Arabic News Sentences'
size_categories:
- 10K<n
dataset_info:
- config_name: ThatiAR
splits:
- name: train
num_examples: 2558
- name: dev
num_examples: 373
- name: test
num_examples: 742
- config_name: ThatiAR-Instruct
splits:
- name: train
num_examples: 2558
- name: dev
num_examples: 373
- name: test
num_examples: 742

configs:
- config_name: ThatiAR
data_files:
- split: train
path: data/subjectivity_2024_train.tsv
- split: dev
path: data/subjectivity_2024_dev.tsv
- split: test
path: data/subjectivity_2024_test.tsv
- config_name: ThatiAR-Instruct
data_files:
- split: train
path: instruction_explanation_dataset/subjectivity_2024_instruct_train.json
- split: dev
path: instruction_explanation_dataset/subjectivity_2024_instruct_dev.json
- split: test
path: instruction_explanation_dataset/subjectivity_2024_instruct_test.json
---

# ThatiAR: Subjectivity Detection in Arabic News Sentences

Along with the paper, we release the dataset and other experimental resources. Please find the attached directory structure below.

### Files Description

- **data/**
- `subjectivity_2024_dev.tsv`: Development set for subjectivity detection in Arabic news sentences.
- `subjectivity_2024_test.tsv`: Test set for subjectivity detection in Arabic news sentences.
- `subjectivity_2024_train.tsv`: Training set for subjectivity detection in Arabic news sentences.

- **instruction_explanation_dataset/**
- `subjectivity_2024_instruct_dev.json`: Development set with instruction explanations.
- `subjectivity_2024_instruct_test.json`: Test set with instruction explanations.
- `subjectivity_2024_instruct_train.json`: Training set with instruction explanations.

- `licenses_by-nc-sa_4.0_legalcode.txt`: License information for the dataset, under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
- `README.md`: This readme file containing information about the dataset and its structure.

## License

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. You can view the full license in the `licenses_by-nc-sa_4.0_legalcode.txt` file.

## Usage

To use this dataset, you can load the TSV or JSONL files into your data processing pipeline.

### Example (Python)

```python
import pandas as pd
import json

def load_tsv(file_path):
return pd.read_csv(file_path, sep='\t')

def load_json(file_path):
with open(file_path, 'r', encoding='utf-8') as file:
data = json.load(file) # Use json.load() for reading standard JSON files
return data

# Load training data
train_data_tsv = load_tsv('data/subjectivity_2024_train.tsv')
train_data_jsonl = load_json('instruction_explanation_dataset/subjectivity_2024_instruct_train.json')

# Load development data
dev_data_tsv = load_tsv('data/subjectivity_2024_dev.tsv')
dev_data_jsonl = load_json('instruction_explanation_dataset/subjectivity_2024_instruct_dev.json')

# Load test data
test_data_tsv = load_tsv('data/subjectivity_2024_test.tsv')
test_data_jsonl = load_json('instruction_explanation_dataset/subjectivity_2024_instruct_test.json')
```

### Data splits
We split the dataset in a stratified manner, allocating 70\%, 10\%, and 20\% for training, development, and testing, respectively.

## Citation
```
@article{ThatiAR2024,
title = {{ThatiAR}: Subjectivity Detection in Arabic News Sentences},
author = {Suwaileh, Reem and Hasanain, Maram and Hubail, Fatema and Zaghouani, Wajdi and Alam, Firoj},
year = {2024},
journal = {arXiv: 2406.05559},
}
```