IbraahimLab/fineweb-somali

FineWeb Somali Dataset

A curated collection of Somali language content from BBC Somali, designed for training and evaluating small language models on low-resource languages.

Dataset Description

This dataset contains 4,910 high-quality Somali language articles scraped from BBC Somali's website. The content covers diverse topics including news, culture, technology, sports, and human interest stories, providing a rich corpus for Somali language model training.

Dataset Summary

• Language: Somali (so)
• Total Records: 4,910 articles
• Source: BBC Somali (bbc.com/somali)
• Collection Date: January 25, 2026
• License: MIT (for educational and research purposes only)
• Purpose: Educational - for building and experimenting with small language models for low-resource languages

Key Features

• ✅ High-Quality Content: Professional journalism from BBC Somali
• ✅ Diverse Topics: News, sports, culture, technology, health, and more
• ✅ Clean Text: Pre-processed and validated Somali text
• ✅ Metadata Rich: Includes URLs, dates, file paths, and token counts
• ✅ Language Verified: All content has language_score of 1.0

Dataset Structure

Data Fields

Field	Type	Description
text	string	Main article content (136-26,800 characters)
id	string	Unique identifier (UUID format, 36 chars)
dump	string	Dataset dump identifier (CC-MAIN-2026-SOMALI)
url	string	Source article URL (35-431 characters)
date	timestamp	Collection timestamp (2026-01-25 16:56:04)
file_path	string	Source file path (238 unique sources)
language	string	Language code (always "so")
language_score	float	Language detection confidence (always 1.0)
token_count	int	Number of tokens per article (17-4,160 tokens)

Data Splits

Split	Records
train	4,910

Usage

Loading the Dataset

from datasets import load_dataset

# Load the full dataset
dataset = load_dataset("IbraahimLab/fineweb-somali")

# Access the training split
train_data = dataset['train']

# Example: Print first article
print(train_data[0]['text'][:500])

Example Use Cases

1. Training a Small Language Model

from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer

# Load dataset
dataset = load_dataset("IbraahimLab/fineweb-somali")

# Initialize tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("your-base-model")
model = AutoModelForCausalLM.from_pretrained("your-base-model")

# Your training code here...

2. Text Analysis

import pandas as pd

# Convert to pandas for analysis
df = dataset['train'].to_pandas()

# Analyze token distribution
print(df['token_count'].describe())

# Check content sources
print(df['file_path'].value_counts())

3. Data Exploration

# Filter by token count
long_articles = dataset['train'].filter(lambda x: x['token_count'] > 1000)

# Sample random articles
sample = dataset['train'].shuffle(seed=42).select(range(10))

Dataset Statistics

• Total Articles: 4,910
• Total Characters: ~40M (estimated)
• Average Article Length: ~8,150 characters
• Token Count Range: 17 - 4,160 tokens
• Unique Sources: 238 file paths
• Language Purity: 100% (all articles scored 1.0)

Source Information

All content is sourced from BBC Somali (https://www.bbc.com/somali), which provides:

• Professional journalism in Somali language
• Wide topic coverage (news, sports, culture, tech, health)
• High-quality, edited content
• Diverse vocabulary and writing styles

Content Categories

Based on file paths, content includes articles from various BBC Somali journalists and sections covering:

• Breaking news and current events
• Sports coverage
• Cultural stories
• Technology and science
• Health and lifestyle
• Human interest stories

Intended Use

Primary Use Cases

• 🎓 Educational: Learning to build language models for low-resource languages
• 🔬 Research: Somali NLP research and experimentation
• 🤖 Model Training: Pre-training or fine-tuning small language models
• 📊 Linguistic Analysis: Studying Somali language patterns and structures

Out-of-Scope Use

• ❌ Commercial applications without proper licensing
• ❌ Creating misinformation or fake news in Somali
• ❌ Any use that violates BBC's terms of service

Limitations

• Single Source: All content is from BBC Somali only
• Temporal Scope: Single collection date (January 25, 2026)
• Domain: Primarily news/journalism style, may not represent conversational Somali
• Size: Moderate dataset, may not be sufficient for large model training
• License: Educational/research purposes only

Ethical Considerations

• This dataset is provided for educational and research purposes only
• Original content belongs to BBC and should be used in accordance with fair use principles
• Users should respect copyright and only use this for non-commercial learning purposes
• Researchers should cite both this dataset and BBC Somali as the original source

Citation

If you use this dataset, please cite:

@dataset{fineweb_somali_2026,
  author = {IbraahimLab},
  title = {FineWeb Somali: BBC Somali Content for Language Model Training},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/IbraahimLab/fineweb-somali}
}

Also acknowledge the original source:

Content sourced from BBC Somali (https://www.bbc.com/somali)

Dataset Creation

Collection Process

• Method: Web scraping from BBC Somali website
• Date: January 25, 2026
• Processing: Text extraction, cleaning, and validation
• Quality Control: Language detection (all articles scored 1.0 for Somali)

Preprocessing

• Text extracted from HTML content
• Character encoding standardized
• Token counts computed
• Metadata preserved (URLs, dates, sources)

Maintenance

• Version: 1.0
• Last Updated: January 2026
• Status: Static dataset (snapshot from single date)

Contact

For questions or issues regarding this dataset:

• Creator: IbraahimLab
• Platform: Hugging Face Datasets

Acknowledgments

• BBC Somali: Original content creators
• Somali Language Community: For preserving and developing the language
• Low-Resource Language Research Community: For supporting Somali NLP development

---

Disclaimer: This dataset is intended solely for educational and research purposes. All content rights belong to BBC. Users must comply with applicable copyright laws and BBC's terms of service.