Thank you.
I would like to say thank you for this
Although processing this dataset is probably going to be a nightmare.
It seems the brazilians may have not taken into account to keep the author names/ids consistent. I put a select portion of a guild's messages into a LLM, and in it's thinking it literally pointed out that there's someone with a different name but same id.
I would also like to inquire about whether or not you have anonymized usernames, and/or ids.
I would also like to inquire about whether or not you have anonymized usernames, and/or ids.
I have not applied any post processing to the dataset I uploaded, it's a clone so people can get much better download speeds ^_^
Anonymization details in the paper claim
- Usernames: Replaced with consistent pseudonyms (using the mimesis library).
- User IDs & Message IDs: Hashed using SHA-256 and truncated to 12 characters, maintaining linkage while masking originals.
global_namefield: Entirely removed.- User IDs in Content: Identified via regular expressions within message text and replaced with their corresponding hash values.
From my peek into the data, take 4 with a grain of salt :3