Japanese conversational data for AI training and research
Structured access to historical and continuously updated public discussions from 5channel — one of Japan's largest and longest-running online communities.
Licensed directly by the operator of 5channel.
Native Japanese at scale
Informal, conversational Japanese as actually written: slang, abbreviations, kaomoji, board-specific register. Text of a kind that is underrepresented in web-crawled corpora.
Decades of history
Continuous archives spanning the modern history of the Japanese internet, plus ongoing daily additions.
Full thread structure
Not flat text. Threads, posts, reply and quotation relationships, board classification, and timestamps are preserved.
Configured to your requirements
Filter by board, date range, thread length, or content type. Delivery format and update cadence are set per project.
Documented provenance
Licensed directly from the operator, with a documented chain of custody suitable for enterprise and public-sector diligence.
Standard formats
JSONL and Parquet as standard; other formats and delivery methods available by arrangement.
Why license rather than crawl
Completeness.
Crawling yields a partial and structurally degraded copy — truncated threads, lost reply relationships, missing historical depth, and rate-limited coverage. We deliver the underlying records with structure intact.
Provenance.
Models entering public-sector procurement, regulated industries, or enterprise deployment increasingly require a documented data supply chain. A licence executed with the source operator is the strongest form of that documentation.
Scope beyond training.
Retrieval, search, summarisation, and answer-generation products use source material in ways that differ from model training. Licensed access defines those permissions explicitly.
What the corpus contains
5channel is a large-scale Japanese anonymous discussion board organised into topical boards, each containing threads, each thread containing sequentially numbered posts. The corpus we license is the public post record of that system, together with the structural metadata that gives it meaning.
The corpus is dense in informal spoken-register Japanese, internet slang, kaomoji, and multi-party threaded exchange — training signal that is difficult to source elsewhere at comparable scale or continuity.
{
"post_id": "5ch:news4vip:1712345678:0042",
"thread_id": "5ch:news4vip:1712345678",
"board_id": "news4vip",
"board_name_ja": "ニュー速VIP",
"board_category_ja": "雑談系",
"post_index": 42,
"posted_at": "2024-04-05T21:14:38+09:00",
"posted_at_precision": "second",
"name_field": "以下、5ちゃんねるからVIPがお送りします",
"poster_id": "aB3xY9Zk0",
"body_text": "そういうのはこっちのスレでやってくれ\n>>38 が言ってたやつな",
"body_normalized": "そういうのはこっちのスレでやってくれ >>38 が言ってたやつな",
"reply_to": [38],
"quoted_spans": [
{ "target_post_index": 38, "raw": ">>38" }
],
"char_count": 34,
"contains_ascii_art": false,
"redaction_profile": "standard",
"redactions_applied": 0,
"source_snapshot_id": "snap-2026-08-01",
"record_version": 1
}Request a dataset sample.
We provide a representative technical sample, schema documentation, and a dataset brief to qualified organizations.