Mislabelled polarities relative to original datasets

[asmith-wowcorp]

I was curious because there seemed like many mis-labelled examples in the acos datasets, so I compared one I found with another source for the dataset. In InstructABSA repo, I can see:

whereas in this repo the second sentiment for that comment is positive: https://github.com/yangheng95/ABSADatasets/blob/v2.0/datasets/acos_datasets/504.Restaurant16/rest16_quad_train.tsv.txt

It has me bit worried about using this source :-/. Any idea what is going on?

[asmith-wowcorp]

It looks like the source of the ACOS annotations is here: https://github.com/NUSTM/ACOS/tree/main which has the wrong polarity already