OwnedByDanes/Usenet-Corpus-1980-2013

Usenet Corpus 1980–2013

One of the largest curated Usenet archives available for AI training — 103.1 billion tokens of authentic pre-LLM human discourse.

This dataset contains 408 million cleaned and deduplicated Usenet posts spanning 1980–2013 across 18,347 newsgroups. It is sourced from one of the largest privately held Usenet corpora and has been rigorously processed for modern AI training use cases.

⚠️ Note on size indicator: The size badge above reflects the sample files hosted in this repository (65K rows). The full corpus is 103.1B tokens across 408M posts. See Key Statistics below for complete figures.

Sample files are free to download with no approval required. The full corpus is available for licensing — see Licensing & Access below.

⚠️ Correction (May 28, 2026): An earlier version of the sample files contained unhashed Message-IDs in the id field. These have been corrected — all id values are now SHA-256 hashed (msg-<hex32>). The full corpus cleaning pipeline is being updated accordingly. We thank the community for identifying this issue.

---

Key Statistics

Stat	Value
Total posts	408,236,288
Total tokens	103.1 billion (cl100k_base)
Overall drop rate	0.11% (450,990 records filtered)
Primary language	English (96.6%)
Newsgroups	18,347
Date range	1980–2013

Major Hierarchies by Tokens

Hierarchy	Tokens	Posts	Notes
alt.*	58.8B	228.8M	Highest volume and diversity
rec.*	16.5B	77.2M	Recreational topics
comp.*	10.3B	50.0M	Technical depth
soc.*	8.2B	22.4M	Most multilingual
sci.*	3.3B	12.0M	Scientific and research-grade

---

Models Trained on This Dataset

This corpus is already being used in real training projects:

• wyan/usenet-gemma-4-E2B-lora
• wyan/usenet-gemma-4-E2B-lora-it (Italian variant)

Early feedback indicates that the comp.* and sci.* hierarchies provide significantly cleaner signal than alt.*, and that temporal filtering (e.g., pre- vs post-1995) can further improve model output quality.

---

Data Fields

Each record includes:

Field	Type	Description
text	string	Post body (cleaned)
group	string	Newsgroup name
date	string	ISO 8601 timestamp where available
subject	string	Thread subject
author	string	Display name (email addresses stripped)
id	string	Anonymized Message-ID (SHA-256 hashed to msg-<hex32>). Corrected May 28, 2026 — earlier sample versions contained unhashed IDs.

---

Data Quality & Processing

The corpus has undergone extensive two-stage cleaning:

Stage 1 — Hierarchy-level exclusions

• Removed alt.binaries.* entirely (UUencoded binary content)
• Removed adult content newsgroups

Stage 2 — Record-level cleaning

• Deduplication by Message-ID
• Binary removal (UUencode and Base64 detection via MIME header inspection)
• PII redaction — email addresses replaced with [email] token
• Message-ID anonymization — SHA-256 hashed to msg-<hex32>
• Sensitive content filtering (spam, bulk PII patterns)

Main drop reasons:

• Sensitive content: 232,626
• UUencode lines: 126,441
• Base64 content: 75,164
• UUencode blocks: 16,315

99% of files validated clean with zero email or PII violations in final scan.

---

Intended Uses

• Pre-training and continued pre-training of LLMs, under a licensing agreement
• Domain adaptation for technical, scientific, and historical internet discourse
• Dialogue and long-context modeling
• Linguistic research and digital humanities
• Debiasing and temporal language studies
• Evaluation and experimentation using the freely available sample files

Not intended for: Re-identification of individuals, automated profiling, or spam generation.

---

Licensing & Access

This dataset is released under a custom license. Terms differ between sample files and the full corpus:

Sample files (available in this repository):

• Free to download — no approval required
• Permitted for evaluation, non-commercial research, and personal experimentation
• Non-commercial fine-tuning on sample files is permitted

Full corpus (103.1B tokens, 408M posts):

• Available under a separate written licensing agreement
• Academic research licenses available
• Commercial AI training licenses available — contact for pricing
• Redistribution not permitted without written agreement

To discuss licensing or data partnerships: Contact OwnedByDanes via HuggingFace, open a Discussion on this page, or email usenetoverlord@gmail.com.

---

Provenance & Derivative Work

The underlying posts are public Usenet communications originally published in publicly accessible newsgroups. The compiled, processed corpus represents a derivative work produced through substantial original effort — format conversion, deduplication, sanitization, binary removal, PII redaction, and curation — which distinguishes it from raw archival source material.

---

Citation

@dataset{usenet_corpus_1980_2013,
  author    = {OwnedByDanes},
  title     = {Usenet Corpus 1980--2013},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/OwnedByDanes/Usenet-Corpus-1980-2013}
}

---

Contact

For commercial licensing, evaluation access, research collaboration, or questions about the dataset:

📩 usenetoverlord@gmail.com

Or open a Discussion on this Hugging Face page.