--- language: - ja pretty_name: 2ch.sc --- # **2ch.sc Corpus** ### A Large-Scale Japanese Anonymous Web Forum Dataset This dataset is a large-scale corpus of user-generated text scraped from `2ch.sc`, a massive and influential anonymous Japanese textboard. As a major successor to the original 2channel (`2ch.net`), it represents a significant, unfiltered snapshot of Japanese internet culture, dialogue, and public opinion spanning decades. The data consists of billions of posts across thousands of topic-specific boards, offering an unparalleled resource for studying informal language, online subcultures, and social trends. ## Dataset Structure The corpus is organized into compressed JSON Lines files (`.jsonl.zst`), where each file represents a single forum board. **File Naming Convention:** The filenames follow the pattern `newdump-{region}_{board_name}.jsonl.zst`. * **`region`**: Corresponds to the server or subdomain (e.g., `ikura`). * **`board_name`**: The name of the specific board (e.g., `anime2`). Each line within a file is a JSON object representing a single thread. ### Data Fields | Field | Type | Description | | :--- | :--- | :--- | | `id` | Integer | The unique identifier for the thread (timestamp-based). | | `title` | String | The title of the thread. | | `region` | String | The server/subdomain where the thread originated. | | `city` | String | The name of the board. | | `replies` | Integer | The total number of replies in the thread. | | `messages` | Array[Object] | An array containing all posts within the thread. | ### Message Object Structure Each object in the `messages` array contains the following fields for an individual post: | Field | Type | Description | | :--- | :--- | :--- | | `name` | String | The username of the poster, often including tripcodes or other identifiers. | | `mail` | String | The content of the "mail" field, often used for names such as `sage`. | | `dateid` | String | Timestamp and user ID string (e.g., `2016/05/19(木) 00:03:16.25 ID:uDE3xZkb0.net`). | | `body` | String | The raw HTML content of the post. | | `title` | String / Null | The title of the post, typically only present for the first post in a thread. |
Click to view a sample JSON object for a single thread ```json { "id": 1463583796, "title": "ガールズ＆パンツァー GIRLS und PANZER 1591輌目", "region": "ikura", "city": "anime2", "replies": 1005, "messages": [ { "name": "名無しさん＠お腹いっぱい。＠＼(^o^)／(ﾜｯﾁｮｲ f8f1-xKv1)", "mail": "hageteon", "dateid": "2016/05/19(木) 00:03:16.25 ID:uDE3xZkb0.net", "body": " !extend:on:vvvvv:1000:512
↑冒頭にコレを二行重ねてスレ立てしておく

戦車道は大和撫子のたしなみ！戦車（パンツァー）のある学園生活、はじめました。
===========================重要===========================================
・【※実況厳禁】→アニメ特撮実況板：ttp://hayabusa.2ch.net/liveanime/
・公式以外の動画サイト、ファイル共有に関する話題・URL貼りは厳禁
・sage進行推奨／煽り、荒らしは徹底放置／2chブラウザの導入推奨
・次スレは>>900が宣言してから立てる事。無反応なら>>930以降でスレ立て宣言順に立てる事
・次スレはwikiからコピペして立てて下さい
・巡礼報告はプライバシーに十分配慮し、投稿前に内容を再確認。ガルパンファンとしてマナーを守りましょう
==========================================================================
●放送／配信情報　平成24年10月8日より　12話放送済
TVシリーズ配信　ttp://girls-und-panzer.jp/mv_distribution.html
●前スレ
ガールズ＆パンツァー GIRLS und PANZER 1590輌目
ttp://hanabi.2ch.net/test/read.cgi/anime2/1463498901/
VIPQ2_EXTDAT: default:vvvvv:1000:512:----: EXT was configured ", "title": "ガールズ＆パンツァー GIRLS und PANZER 1591輌目" }, { "name": "名無しさん＠お腹いっぱい。＠＼(^o^)／ (ﾜﾝﾐﾝｸﾞｸ MMd0-MTTD)", "mail": "sage", "dateid": "2016/05/19(木) 00:10:40.89 ID:QtmQIsphM.net", "body": " >>1
ｱﾘｶﾞﾀｳ、ｱﾘｶﾞﾀｳ ", "title": null } ] } ```
## Data Curation and Processing This dataset is a direct conversion from the native 2ch/5ch `.dat` format to JSON Lines. The raw data was compressed with Zstandard due to its large size (approximately **1.8 TB** uncompressed). The scraping code used for collection is provided in `KakoIndex.py`. ## Language * **Primary Language:** Japanese (ja-JP) * **Characteristics:** The language is highly informal and specific to the platform's culture, containing extensive internet slang, neologisms, and non-standard linguistic phenomena. ## Licensing Information The licensing for this dataset is complex and not clearly defined for academic or research use. The terms of use from the original source material are translated below. The author's interpretation is that the license is functionally similar to **CC-BY-NC-ND** (Attribution-NonCommercial-NoDerivatives). * **1. "Print Out / Copy / Free Distribution" OK Mark** * **Permitted:** Printing, copying, and free distribution of the work *as is*. * **Commercial Use:** Permitted only if the distribution is free (e.g., in a free company pamphlet). * **NOT Permitted:** Any modification, including changing, altering, processing, cutting, partial use, summarizing, translating, transforming, or adapting the work. * **2. "Non-profit Use for People with Disabilities" OK Mark** * **Permitted:** All forms of non-profit use (copying, transmitting, distributing) intended for people with disabilities. This includes modifications and adaptations. * **3. "Non-profit Use for School Education" OK Mark** * **Permitted:** All forms of non-profit use (copying, transmitting, distributing) for school activities. This includes modifications and adaptations. > **Disclaimer:** Researchers should consult the original terms on 2ch.sc and use this data at their own risk, considering the legal and ethical implications. ## Potential Use Cases * **Large Language Model Training:** Training or fine-tuning models on informal, conversational Japanese. * **Sociolinguistics:** Studying the evolution of slang, memes, and online discourse. * **Sentiment Analysis:** Gauging public opinion on various topics (with awareness of significant demographic bias). * **Toxicity and Hate Speech Detection:** Developing models to identify harmful content in Japanese. * **Cultural Analytics:** Tracking and analyzing social trends and subcultural dynamics. ## ⚠️ Ethical Considerations and Limitations This dataset must be handled with extreme care due to its unfiltered and anonymous nature. * **Hate Speech and Offensive Content:** The corpus contains a high volume of explicit, offensive, and discriminatory content. * **Misinformation and Disinformation:** The platform is a significant source of rumors, conspiracy theories, and deliberate misinformation. * **Toxicity and Pervasive Bias:** The data reflects strong biases (e.g., misogynistic, xenophobic, political). Models trained on this data will inherit and likely amplify these biases if not carefully mitigated. * **Lack of Representativeness:** The user base is a self-selecting demographic and is **not representative** of the general Japanese population. * **Privacy Risks:** Although pseudonymous, posts may contain personally identifiable information (PII) through doxxing or self-disclosure. Rigorous PII scrubbing is recommended. ## A Note on the Crawling Process & お詫び (Apology) The original site prohibits crawling without prior consultation. The following note is from the website: > *負荷が掛かるので、事前相談の無いクローリングも禁止です。* > > *Crawling without prior consultation is also prohibited as it places a load on the system.* Although prior consultation was not performed, the crawl was limited to a maximum of 128 concurrent connections, which in practice saturated the connection at approximately 100Mbps to minimize server load.