Pre-training text representations have led to significant improvements in many areas of natural language processing. The quality of these models benefits greatly from the size of the pretraining corpora as long as its quality is preserved. In this paper, we describe an automatic pipeline to extract massive high-quality monolingual datasets from Common Crawl for a variety of languages. Our pipeline follows the data processing introduced in fastText (Mikolov et al., 2017; Grave et al., 2018), that deduplicates documents and identifies their language. We augment this pipeline with a filtering step to select documents that are close to high quality corpora like Wikipedia.
CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data
An automatic pipeline extracts high-quality monolingual datasets from Common Crawl to improve pre-trained text representations in natural language processing.
- Year
- 2019
- Venue
- ccnet-extracting-high-quality-monolingual-1
- Authors
- 7
- Hosting
- Abstract onlyARXIV-DEFAULT
Cite
Notes
Only stored in your browser.
Attribution
- Abstract & full text
- arxiv.org/abs/1911.00359v2ARXIV-DEFAULT
- TL;DR
- Semantic Scholar