Update README.md
Browse files
README.md
CHANGED
|
@@ -5,4 +5,15 @@ language:
|
|
| 5 |
pipeline_tag: text-classification
|
| 6 |
---
|
| 7 |
|
|
|
|
| 8 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 5 |
pipeline_tag: text-classification
|
| 6 |
---
|
| 7 |
|
| 8 |
+
Fine-tuned model for detecting instances of offensive language in Ducth tweets. The model has been trained with [DALC v2.0 ](https://github.com/tommasoc80/DALC).
|
| 9 |
|
| 10 |
+
Offensive language defintion is inhereted from SemEval 2019 OffensEval: "Posts containing any form of non-acceptable language (profanity) or a targeted offence,
|
| 11 |
+
which can be veiled or direct. This includes insults, threats, and posts containing profane language or swear words." ([Zampieri et al., 2019](https://aclanthology.org/N19-1144/))
|
| 12 |
+
|
| 13 |
+
The model achieve the following results on multiple test data:
|
| 14 |
+
|
| 15 |
+
- DALC held-out test set: macro F1: 79.93; F1 Offensive: 70.34
|
| 16 |
+
- HateCheck-NL (functional benchmark for hate speech): Accuracy: 61.40; Accuracy non-hateful tests: 47.61 ; Accuracy hateful tests: 68.86
|
| 17 |
+
- OP-NL (dynamyc benchmark for offensive language): macro F1: 73.56
|
| 18 |
+
|
| 19 |
+
More details on the training settings and pre-processind are available [here](https://github.com/tommasoc80/DALC)
|