drelhaj/AraFinNews

---
language: ar
license: cc-by-4.0
tags:
- arabic
- finance
- summarization
- dataset
task_categories:
- summarization
size_categories:
- 100K+
---

# AraFinNews: The Arabic Financial News Dataset (212K)

## For the JSON file format please check our [AraFinNews GitHub repo](https://github.com/ArabicNLP-UK/AraFinNews)

**AraFinNews** is the largest openly available dataset of Arabic financial news, comprising **212,500 full-length articles** collected from [Argaam.com](https://www.argaam.com/) — a leading financial news portal in the Arab world.
The dataset provides structured, machine-readable text suitable for research in **financial NLP**, **abstractive summarisation**, **event extraction**, and **domain-specific language modelling**.

Comparable to the **CNN/DailyMail** dataset for English, AraFinNews offers an Arabic equivalent for headline-style abstractive summarisation. Each record pairs a full Arabic financial article with its professionally written headline, enabling high-quality training and evaluation of summarisation and financial text understanding systems.

---

## 📰 Dataset Overview

| Field | Description |
|---------|-----------------------------------------------|
| `id` | Unique numeric identifier |
| `title` | Arabic headline (financial news title) |
| `date` | Publication date (ISO format) |
| `article` | Full article text |
| `url` | Public link on *Argaam.com* |

- **Total articles:** 212,500
- **Language:** Modern Standard Arabic
- **Domain:** Finance, markets, economics, corporate activity
- **Format:** CSV (UTF-8)
- **Licence:** CC BY-NC 4.0

---

## ⚙️ Updated Data Splits

In addition to the original ID lists, the repository now includes **fully populated split files**:

- `AraFinNews_train.csv` — 80% of the dataset
- `AraFinNews_validation.csv` — 10%
- `AraFinNews_test.csv` — 10%

These files contain **the complete article and headline rows**, making them directly compatible with the Hugging Face Dataset Viewer and eliminating the need to reconstruct splits manually.

The older ID-only files are still included for reference:

- `AraFinNews_train_ids.csv`
- `AraFinNews_val_ids.csv`
- `AraFinNews_test_ids.csv`

These can be used if a user prefers to work from the master file `AraFinNews.csv`.

---

## ⚙️ Intended Use

AraFinNews supports research in:

- Abstractive and extractive summarisation
- Financial event and entity extraction
- Sentiment and stance analysis in financial narratives
- Domain-specific pretraining and adaptation of Arabic LLMs
- Financial question answering and narrative analysis

The dataset is released strictly for **non-commercial research and educational use**.

---

## 📦 Access and Usage

### Load the full dataset

import pandas as pd

df = pd.read_csv("AraFinNews.csv")
df.sample(5)

### Load directly from the full splits

import pandas as pd

train = pd.read_csv("AraFinNews_train.csv")
val = pd.read_csv("AraFinNews_validation.csv")
test = pd.read_csv("AraFinNews_test.csv")

### (Optional) Load splits using ID files

import pandas as pd

df = pd.read_csv("AraFinNews.csv")
train_ids = pd.read_csv("AraFinNews_train_ids.csv")["id"]
train_df = df[df["id"].isin(train_ids)]

---

## 📚 Citation

If you use this dataset, please cite:

> **El-Haj, M.** & **Rayson, P.** (2025). *AraFinNews: Arabic Financial Summarisation with Domain-Adapted LLMs.* Proceedings of IEEE Big Data 2025.

This work is associated with the following preprint:
**https://arxiv.org/abs/2511.01265**

---

## 🏛️ Repository Structure
Note: json files are on [AraFinNews GitHub](https://github.com/ArabicNLP-UK/AraFinNews)

AraFinNews/
│
├── AraFinNews.csv
├── train.csv
├── validation.csv
├── test.csv
│
├── AraFinNews_train_ids.csv
├── AraFinNews_val_ids.csv
├── AraFinNews_test_ids.csv
│
├── AraFinNews_json_files/
│ ├── 000001.json
│ ├── 000002.json
│ └── ...
│
└── README.md

---

**Contact:**
Mo El-Haj
VinUniversity, Hanoi / Lancaster University, UK
[e-mail]