Question on the data labeling
#2
by
SengTak
- opened
Firstly, I would like to say your work is commendable. But I would like to raise some questions on the annotation of the dataset, was it done internally? Has it undergone review?
One of the data point I came across "that s really sibay jialat" was labeled as "harassment" but I would say most Malaysian wouldn't agree to that...
Hello! the annotation of the dataset involves use of pseudo labeling using LLM, active learning and filtering. You may read about it here https://arxiv.org/pdf/2407.20729.
Since it is not manually labelled, we understand and aware that they are false labels in the dataset.
aisyahhrazak
changed discussion status to
closed