understanding file naming
#3
by nastyafilippova - opened
Thanks for releasing the data! I have a question regarding the shard naming convention. What logic determines the file ranges per topic and the varying shard counts for different prefixes (like common_crawl-industrial 017 vs. 018)?
Hi @nastyafilippova , thanks for the question! The topics (e.g. industrial, literature) are from a fasttext classifier used to parition topics based on the weborganizer paper. The 17 vs 18 is a quality bucket (in ordered buckets of 5-percentile), where quality score was generated from a fasttext quality classifier. Different sizes are because we tried to keep the number of documents the same in each bucket.