🚩 Report: Ethical issue(s)

#2
by Steevo75 - opened

I find it a bit unfortunate that this datased is a bit too politically oriented : 595 occurrences listing hate speech against "gauchistes", two occurrence for "droitards", which is curious, especially since the source is Twitter. No target with "right_wing_people" compared to 383 occurrences for the target "left_wing_people". ethically, all political parties should be treated the same

Steevo75 changed discussion status to closed
Steevo75 changed discussion status to open

Hi @Steevo75 , thanks for your interest! I built this dataset by concatenating all existing annotated resources for French hate speech detection from the academic literature, so your observation means that research has focused more so far on hate speech against left-wing than right-wing. It could also be that hate speech against left-wing is more prevalent than against right-wing in the French-speaking context (I have preliminary evidence that this is the case on French Twitter, will share the dataset supporting this claim on my Hugging Face in a couple of months when my analysis is finished). Finally, you are very welcome to contribute more resources on hate speech against right-wing, I'd be happy to add that to my dataset. Let me know if I addressed your concern or if you have any other questions, thanks!

Hi @manueltonneau , thanks for your answer.
I completely understand your approach, my remark was perhaps not relevant because this type of speech is found in all parties, I would always be able to adjust if necessary.
Your dataset already includes a very large number of data to be used to train a model at RLHF, many thanks to you and sorry for the report.

Steevo75 changed discussion status to closed

Sign up or log in to comment