merrybabyxmas/mindcast-sarcasm-detector

Mindcast Sarcasm Detector

Model Description

한국어 텍스트가 비꼬는 표현(sarcasm)인지 아닌지를 판별하는 이진 분류 모델입니다.

이 모델은 LoRA (Low-Rank Adaptation)를 사용하여 효율적으로 파인튜닝되었으며, 최종적으로 base model과 merge되어 배포되었습니다.

Training Date: 2025-12-15

Performance

Test Set Results

Metric	Score
Accuracy	0.6519
F1 Score (Macro)	0.6499
F1 Score (Weighted)	0.6526

Confusion Matrix

[[196 102]
 [ 86 156]]

Detailed Classification Report

               precision    recall  f1-score   support

Non-sarcastic     0.6950    0.6577    0.6759       298
    Sarcastic     0.6047    0.6446    0.6240       242

    micro avg     0.6519    0.6519    0.6519       540
    macro avg     0.6498    0.6512    0.6499       540
 weighted avg     0.6545    0.6519    0.6526       540

Training Details

Hyperparameters

Hyperparameter	Value
Base Model	klue/roberta-base
Batch Size	64
Epochs	30
Learning Rate	0.0001
Warmup Ratio	0.1
Weight Decay	0.01
LoRA r	8
LoRA alpha	16
LoRA dropout	0.05

Training Data

• Train samples: 1428
• Valid samples: 159
• Test samples: 540
• Number of labels: 2
• Labels: Non-sarcastic, Sarcastic

Usage

Installation

pip install transformers torch

Quick Start

from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline

# Load model
model_name = "merrybabyxmas/mindcast-sarcasm-detector"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

# Create pipeline
classifier = pipeline("text-classification", model=model, tokenizer=tokenizer)

# Predict
text = "아 진짜 좋네요~ 완전 최고에요~"
result = classifier(text)
print(result)
# Output: [{'label': 'Sarcastic', 'score': 0.92}]

Model Architecture

• Base Model: klue/roberta-base
• Task: Sequence Classification
• Number of Labels: N/A

Citation

If you use this model, please cite:

@misc{mindcast-model,
  author = {Mindcast Team},
  title = {Mindcast Sarcasm Detector},
  year = {2025},
  publisher = {HuggingFace},
  howpublished = {\url{https://huggingface.co/merrybabyxmas/mindcast-emotion-sc-only}},
}

Contact

For questions or feedback, please open an issue on the model repository.

---

This model card was automatically generated.