Data · dataset · 2022
aps/dynahate
Listed in Hugging Face Datasets
We present a human-and-model-in-the-loop process for dynamically generating datasets and training better performing and more robust hate detection models.
Description
We provide a new dataset of ~40,000 entries, generated and labelled by trained annotators over four rounds of dynamic data creation. It includes ~15,000 challenging perturbations and each hateful entry has fine-grained labels for the type and target of hate.
Hateful entries make up 54% of the dataset, which is substantially higher than comparable datasets. We show that model performance is substantially improved using this approach. Models trained on later rounds of data collection perform better on test sets and are harder for annotators to trick.
Read the rest (1 more)
They also perform better on HATECHECK, a suite of functional tests for online hate detection. See arxiv.org/abs/2012.15761 for more details.
Links
Where it is published
- Hugging Face dataset page huggingface.co/datasets/aps/dynahate ↗
landing page · from Hugging Face
Catalogue records · 1
- Hub API huggingface.co/api/datasets/aps/dynahate ↗
metadata API · from Hugging Face
Topics
- From keywords
- Computer Science & AI
Provenance · 1 source records, 7 field assertions
| Source | Key | Last seen | Raw |
|---|---|---|---|
| Hugging Face Datasets | aps/dynahate | 14 d ago | JSON v1 |
| Field | Assertion | Extractor | Evidence |
|---|---|---|---|
| access_level | source · Hugging Face | connector:huggingface@1.0.0 | /gated |
| concepts[field].local:field:computer-science-ai | mapping · Hugging Face | connector:huggingface@1.0.0 | |
| created_date | source · Hugging Face | connector:huggingface@1.0.0 | |
| description | source · Hugging Face | connector:huggingface@1.0.0 | /description |
| publication_date | source · Hugging Face | connector:huggingface@1.0.0 | |
| title | source · Hugging Face | connector:huggingface@1.0.0 | /id |
| updated_date | source · Hugging Face | connector:huggingface@1.0.0 |