Dataset Card for Oasis-Corpus

Dataset Description

Oasis-Corpus is a 783GB high-quality bilingual corpus.

All data in Oasis-Corpus are built by Oasis and sourced from Common Crawl. It consists of 374GB of Chinese from 17 recent dumps and 409GB of English textual data from 5 dumps.

Languages

English(409GB, 70,121,125 lines) and Chinese(374GB, 110,580,964 lines)

Data Splits

Language	Dump	docs	size
Chinese	cc-may-jun-2023-zh	5,627,020	19.31 GB
	cc-mar-apr-2023-zh	5,548,376	19.22 GB
	cc-jan-feb-2023-zh	5,369,296	18.55 GB
	cc-sep-oct-2022-zh	6,156,501	20.86 GB
	cc-aug-2022-zh	4,971,629	17.14 GB
	cc-jun-jul-2022-zh	5,566,643	18.85 GB
	cc-may-2022-zh	6,408,203	21.53 GB
	cc-jan-2022-zh	6,853,895	22.70 GB
	cc-oct-2021-zh	7,975,739	26.35 GB
	cc-sep-2021-zh	7,371,460	24.69 GB
	cc-jul-aug-2021-zh	6,643,794	22.17 GB
	cc-jun-2021-zh	6,509,108	22.25 GB
	cc-may-2021-zh	5,142,078	17.63 GB
	cc-apr-2021-zh	7,284,775	24.32 GB
	cc-jan-2021-zh	8,133,760	27.19 GB
	cc-nov-dec-2020-zh	6,834,254	23.49 GB
	cc-oct-2020-zh	8,184,433	27.40 GB
English	cc-may-jun-2023-en	15,712,655	90.74 GB
	cc-may-2022-en	14,728,252	81.81 GB
	cc-jun-jul-2022-en	14,124,173	81.66 GB
	cc-jan-2022-en	12,686,195	78.67 GB
	cc-oct-2021-en	12,869,850	75.24 GB

Dataset Structure

Data Fields

• text:the processed and cleaned text contained in the page

• timestamp:timestamp of when the webpage was crawled by CommonCrawl

• url:the url of the webpage crawled to produce the sample

Dataset Creation

• (1) Ungoliant Content Extraction

• (2) Rule Filter

• (3) Neural Filter

• (4) Document Deduplication

Contact

The Laboratory of Cognition and Decision Intelligence for Complex Systems. Institute of Automation, Chinese Academy of Sciences

tongzhou21@outlook.com

yubo.chen@nlpr.ia.ac.cn