Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,36 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- ckb
|
| 4 |
+
tags:
|
| 5 |
+
- kurdish
|
| 6 |
+
- sentiment-analysis
|
| 7 |
+
- offensive-language-detection
|
| 8 |
+
- multi-task-learning
|
| 9 |
+
- kubert
|
| 10 |
+
---
|
| 11 |
+
|
| 12 |
+
# KurdSense — Multi-Task Kurdish Sentiment & Offensive Language Detector
|
| 13 |
+
|
| 14 |
+
KurdSense is the first multi-task NLP model for Kurdish Sorani, simultaneously predicting **sentiment** (Negative/Neutral/Positive) and **offensiveness** (Safe/Offensive) from a single Kurdish tweet.
|
| 15 |
+
|
| 16 |
+
## Model Details
|
| 17 |
+
- **Base model:** KuBERT (asosoft/KuBERT-Central-Kurdish-BERT-Model)
|
| 18 |
+
- **Architecture:** Shared KuBERT encoder + two classification heads
|
| 19 |
+
- **Language:** Kurdish Sorani (ckb)
|
| 20 |
+
- **Dataset:** STSD — 24,668 annotated Kurdish tweets (Wady, Badawi & Kurt, 2024)
|
| 21 |
+
|
| 22 |
+
## Results
|
| 23 |
+
|
| 24 |
+
| Task | Macro F1 | Accuracy |
|
| 25 |
+
|---|---|---|
|
| 26 |
+
| Sentiment (3-class) | 0.5100 | 55.9% |
|
| 27 |
+
| Offensiveness (binary) | 0.6326 | 87.1% |
|
| 28 |
+
|
| 29 |
+
## How to Use
|
| 30 |
+
|
| 31 |
+
Load the model weights using the custom KurdSenseMultiTask architecture
|
| 32 |
+
and BertTokenizer with unk_token="[UNK]".
|
| 33 |
+
|
| 34 |
+
## Citation
|
| 35 |
+
Wady, S. H., Badawi, S., & Kurt, F. (2024). A Kurdish Sorani Twitter dataset
|
| 36 |
+
for language modelling. Data in Brief, 57, 110967.
|