CCNet: Extracting High Quality Monolingual Datasets from Web Crawl Data

TL;DR

This paper presents an automated pipeline for extracting large-scale, high-quality monolingual datasets from Common Crawl, enhancing pretraining data quality for natural language processing models.

Contribution

It introduces a novel pipeline that combines deduplication, language identification, and quality filtering to produce superior monolingual datasets from web crawl data.

Findings

Successfully extracted high-quality datasets for multiple languages.

Improved data quality leads to better pretraining model performance.

Pipeline is scalable and adaptable to various languages.

Abstract

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.