Upload folder using huggingface_hub
Browse files- README.md +38 -0
- pytorch_model.bin +3 -0
README.md
ADDED
|
@@ -0,0 +1,38 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- hi
|
| 4 |
+
- en
|
| 5 |
+
tags:
|
| 6 |
+
- emotion-recognition
|
| 7 |
+
- code-mixing
|
| 8 |
+
- hinglish
|
| 9 |
+
- conversation
|
| 10 |
+
- semeval-2024
|
| 11 |
+
datasets:
|
| 12 |
+
- masac
|
| 13 |
+
metrics:
|
| 14 |
+
- f1
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# HingRoBERTa ERC — E-MASAC
|
| 18 |
+
|
| 19 |
+
Fine-tuned [l3cube-pune/hing-roberta](https://huggingface.co/l3cube-pune/hing-roberta)
|
| 20 |
+
for Emotion Recognition in Conversation (ERC) on the SemEval-2024 Task 10 E-MASAC benchmark
|
| 21 |
+
(Hinglish TV dialogues from *Sarabhai vs. Sarabhai*).
|
| 22 |
+
|
| 23 |
+
## Performance
|
| 24 |
+
- **Test wF1: 0.4852** (HingRoBERTa + SVM soft ensemble)
|
| 25 |
+
- Phase 1 alone: wF1 = 0.3901
|
| 26 |
+
- +12.5 pts above SemEval LSTM baseline
|
| 27 |
+
|
| 28 |
+
## Architecture
|
| 29 |
+
Two-phase ensemble:
|
| 30 |
+
- Phase 1: Fine-tuned HingRoBERTa with Focal Loss + R-Drop, speaker embeddings, emotion-conditioned context
|
| 31 |
+
- Phase 2: SVM-RBF on 3,185-d features (Phase 1 embeddings + hand-crafted conversational features)
|
| 32 |
+
|
| 33 |
+
## Labels
|
| 34 |
+
`anger`, `contempt`, `disgust`, `fear`, `joy`, `neutral`, `sadness`, `surprise`
|
| 35 |
+
|
| 36 |
+
## Citation
|
| 37 |
+
Chakradhar Kotha, Konda Bhanu Sai Akhil, Venkata Siva Sai Krishna Pokuri
|
| 38 |
+
IIIT Hyderabad | SemEval-2024 Task 10
|
pytorch_model.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ac3a9c4a8edab45cf8d51f9fb880dcba7b6b5029e1a4d2f409038b547a25bdc7
|
| 3 |
+
size 1115498847
|