Datasets:
is this published dataset finished PII process too?
Hey,
Thanks in advance.
I deeply thank you for the thorough work and documentation including various tips!
My question is that this published dataset is finished PII process too?
As I'm reading the dataset information at the first glance, I thought the PII has done on this dataset using datatrove.
However, when I look the dataset, it seems like pii has been done but several examples(but which is pretty obvious example that needs to be finished) have not been. For instance, below are examples included in the 'CC-MAIN-2016-50' index.
When I install datatrove and apply PIIFormater, then email@example.com has been replaced with 'EMAIL'
"~~ 0813 7176 0777 atau 0852 7897 5248\nvia EMAIL: email@example.com atau\nbisa invite PIN halloriau.com: 2B2591DA\n(mohon dilampirkan data diri Anda)\nKomentar Anda :~~"'When I install datatrove and apply PIIFormater, then JASON@JEEPWORLD.COM has NOT been replaced.
"~~ JW\nWANT TO SHOW YOUR JEEP OFF TO THE WORLD? IF YOU HAVE INTEREST IN YOUR JEEP BECOMING THE FEATURED JEEP OF THE WEEK PLEASE EMAIL JASON@JEEPWORLD.COM\nMay 7, 2013\nMorgan from Mammoth Lakes, CA: “I love how I can get to these incredible plac~~ "
To conclude my understaing, the first case proves that even the obvious example has not been replaced, and the second one proves that it doesn't seem to support capital case. Therefore, excluding second case(which is limitation of the library maybe?) , the first one has not to be appeared if all data has been done with pii.
So it would be great to clarify about the pii process details! thank you
Hi,
we applied this datatrove block to anonymize some PII information: https://github.com/huggingface/datatrove/blob/main/src/datatrove/pipeline/formatters/pii.py. The released dataset had this block applied to it
For your first example:
~~ 0813 7176 0777 atau 0852 7897 5248\nvia EMAIL: email@example.com atau\nbisa invite PIN halloriau.com: 2B2591DA\n(mohon dilampirkan data diri Anda)\nKomentar Anda :~~
In fact "email@example.com" is already a replaced email, in the original text you would have found some other (real) email. Our code will replace emails with one of either "email@example.com" or "firstname.lastname@example.org". If you want matches to be replaced with something else you can change email_replacement="my string"
For the second example it seems that you are correct, our regex does not include uppercase letters. I will update the datatrove code with a fix. Thank you for bringing this to our attention.
Thank you for your kind reply!