Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- tr
|
| 4 |
+
tags:
|
| 5 |
+
- subjectivity
|
| 6 |
+
- CLEF2023
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
Fine-tuned [mDeBERTa V3](https://huggingface.co/microsoft/mdeberta-v3-base) model for subjectivity detection in newspaper sentences.
|
| 10 |
+
This model was developed as part of the CLEF 2023 CheckThat! Lab [Task 2: Subjectivity in News Articles](https://checkthat.gitlab.io/clef2023/task2/).
|
| 11 |
+
|
| 12 |
+
The goal in this task is to detect whether a sentence is objective (OBJ) or subjective (SUBJ). A sentence is subjective if its content is based on or influenced by personal feelings, tastes, or
|
| 13 |
+
opinions. Otherwise, the sentence is objective. [(Antici et al., 2023)](https://ceur-ws.org/Vol-3370/paper10.pdf).
|
| 14 |
+
|
| 15 |
+
The model was fine-tuned using a multilingual training and Turkish development dataset, for which the following (hyper)parameters were utilized:
|
| 16 |
+
```
|
| 17 |
+
Batch Size = 64
|
| 18 |
+
Max Epochs = 2
|
| 19 |
+
Learning Rate = 6e-5
|
| 20 |
+
Warmup Steps = 300
|
| 21 |
+
Weight Decay = 0.1
|
| 22 |
+
```
|
| 23 |
+
|
| 24 |
+
The model ranked first in the CheckThat! Lab and obtained a macro F1 of 0.90 and a SUBJ F1 of 0.91.
|
| 25 |
+
|
| 26 |
+
*DISCLAIMER*: the Turkish data was obtained from Tweets rather than newspaper articles.
|