Usenet-Corpus-1980-2013-Threaded-Samples
收藏资源简介:
Usenet Corpus 1980–2013 — Threaded (Samples) 是一个专为展示和评估而设计的Usenet帖子样本数据集。该数据集源自完整的 Usenet Corpus 1980–2013 — Threaded 语料库(包含约4.079亿条帖子、1.917亿个线程、1030亿个token),本样本仅提供少量可浏览的示例,旨在让用户直观了解完整数据集的线程化结构。数据集共包含11个gzip压缩的JSON Lines文件:每个新闻组层级对应一个文件(每条约5000条帖子,ALT层级约2000条),另外包含一个Big 8合并文件和一个所有层级合并文件(各约9000条帖子)。样本中的帖子被有意选取为至少包含3条帖子的完整线程,以突出展示真实的对话交互,但这一分布并不代表完整语料库的特征(完整语料库中约82%为单帖线程)。每个文件内帖子按线程分组,并按thread_position字段排序,便于阅读重构后的对话。跨帖子重复项已在样本中移除。每条记录包含以下字段:id(稳定的帖子标识符)、group(帖子存档的新闻组)、date(帖子原始日期,RFC 2822格式)、author(作者显示名称,电子邮件地址已编辑)、subject(主题行)、text(帖子正文)、thread_id(会话标识符)、thread_position(帖子在线程中的位置,从0开始)、thread_depth(回复深度,根节点为0)。数据集适用于文本生成(如对话生成)和文本分类(如主题分类、语言识别)等任务。数据经过清理:移除了二进制附件、alt.binaries.* 新闻组、成人内容组,作者电子邮件地址已编辑,Message-ID已替换为哈希,并通过ClamAV恶意软件扫描且未检出。内容保持原始发布状态,未经过滤,可能包含冒犯性、偏见或错误信息,应作为历史原始资料对待。许可证为专有,本样本仅用于评估,不授予任何使用、复制、重新分发或基于数据训练的权利。如需商业许可,需联系OwnedByDanes LLC。
Usenet Corpus 1980–2013 — Threaded (Samples) is a dataset of Usenet post samples designed for demonstration and evaluation. It is derived from the full Usenet Corpus 1980–2013 — Threaded (approximately 407.9 million posts, 191.7 million threads, 103 billion tokens). This sample provides a small set of browsable examples to give users an intuitive understanding of the threaded structure of the full corpus. The dataset consists of 11 gzip-compressed JSON Lines files: one file per newsgroup hierarchy (about 5000 posts each, ALT hierarchy about 2000 posts), plus a Big 8 combined file and an all-hierarchy combined file (about 9000 posts each). Posts are intentionally selected as complete threads containing at least 3 posts to highlight real conversational interactions, but this distribution does not represent the full corpus (where about 82% of threads are single-post). Within each file, posts are grouped by thread and sorted by the thread_position field for easy reading of reconstructed conversations. Cross-post duplicates have been removed. Each record contains the following fields: id (stable post identifier), group (newsgroup where the post was archived), date (original post date, RFC 2822 format), author (display name, email address redacted), subject (subject line), text (post body), thread_id (conversation identifier), thread_position (position of the post within the thread, starting from 0), thread_depth (reply depth, root is 0). The dataset is suitable for text generation (e.g., dialogue generation) and text classification (e.g., topic classification, language identification). Data has been cleaned: binary attachments, alt.binaries.* newsgroups, and adult content groups removed; author email addresses redacted; Message-IDs replaced with hashes; scanned with ClamAV with no detections. Content is kept as originally posted, unfiltered, and may contain offensive, biased, or inaccurate information, and should be treated as historical raw material. License is proprietary; this sample is for evaluation only, and no rights to use, copy, redistribute, or train models are granted. For commercial licensing, contact OwnedByDanes LLC.





