dup_signals missing
Thanks for your great work! I really appreciate the effort you've put into this project. I noticed that the dup_signals in the CommonCrawl part always appears to be empty. However, in your blog post, you mentioned upsampling documents based on the duplicate counts. I wonder if dup_signals should contain the duplicate counts of documents.
Hi, @izusa
Thanks for noticing our work!
Yes, we should have dup_signal for all the data due to the upsampling reasons you mentioned. I haven't checked the full dump yet, these metadata are there for the ones I just sampled.
If the dup_signal are missing, then it might be a mistake on our side. I'd appreciate if you could let us know which file you identified the problem.
But at the meantime, even if the signals are indeed missing, we have organized the documents into folders (such as 101-1000), our upsampling recipe is actually based on this folder structure. But if you need more finegrained study, you'd definitely need the metadata.
P.S. we have incomplete dup_signal that missed the full dup distribution for some files due to our early stage of processing, but that's a separate issue.
Best,
Hector L
Thank you for your response. I hadn't initially realized that the folder names indicate the duplicate counts, and I had only focused on the 1-1. After reviewing it again, I found the dup_signal in the larger duplicate documents.