I think that you are a bit overreacting. Some bias exists probably in every dataset. Now the interesting things is that from what culture roam this dataset wa

I think that you are a bit overreacting. Some bias exists probably in every dataset. Now the interesting things is that from what culture roam this dataset was acquired from considering that operating markets of Yahoo are US, Europe, Japan? Does it resemble demographics of these markets? Or does it represent the desires of these markets? More importantly is there are a good validation set of subtle examples of something being NSFW and SFW? Especially considering that the subject is mostly defined as I do not know to how to define it I definitely know it when I see it.