Datasets:
NSFW processing question
#3
by
grider-withourai
- opened
I'm a bit curious how you dealt with "fewer dramatised duplicated characters (for example 「ああああーーー」)". Would you be able to share your method or script on this, pls? Trying to deal with some NSFW datasets that degrade the performance of the model cause of this. Did you just remove them all?
Hey grider-withourai,
The script simply has a list of characters that tend to be duplicated such as あいうえおー～ and others, then removes any lines where there are 4 or more in a row.
Throughout versions, I've been inconsistent in how I've done this. In next version it should be much better.
I hope that answers your question.