emotion-sphobert: Emotion Recognition for Vietnamese Text

This model is a fine-tuned version of vinai/phobert-base-v2 on the VSMEC dataset for emotion recognition in Vietnamese text.

Model Details

• Base Model: vinai/phobert-base-v2
• Description: SPhoBERT - Specialized PhoBERT with mean pooling
• Dataset: VSMEC (Vietnamese Social Media Emotion Corpus)
• Fine-tuning Framework: HuggingFace Transformers
• Task: Emotion Classification (7 classes)

Hyperparameters

• Batch size: 32
• Learning rate: 2e-5
• Epochs: 100
• Max sequence length: 256
• Weight decay: 0.01
• Warmup steps: 500

Dataset

The model was trained on the VSMEC dataset, which contains 6,927 Vietnamese social media text samples annotated with emotion labels. The dataset includes the following emotion categories:

• Enjoyment (0): Positive emotions, joy, happiness
• Sadness (1): Sad, disappointed, gloomy feelings
• Anger (2): Angry, frustrated, irritated
• Fear (3): Scared, anxious, worried
• Disgust (4): Disgusted, repelled
• Surprise (5): Surprised, shocked, amazed
• Other (6): Neutral or unclassified emotions

Results

The model was evaluated using the following metrics:

• Accuracy: 0.0000
• Macro-F1: 0.0000
• Macro-Precision: 0.0000
• Macro-Recall: 0.0000

Usage

You can use this model for emotion recognition in Vietnamese text. Below is an example of how to use it with the HuggingFace Transformers library:

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("visolex/emotion-sphobert")
model = AutoModelForSequenceClassification.from_pretrained("visolex/emotion-sphobert")

# Example text
text = "Tôi rất vui vì hôm nay trời đẹp!"

# Tokenize
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=256)

# Predict
outputs = model(**inputs)
predicted_class = outputs.logits.argmax(dim=-1).item()

# Map to emotion name
emotion_map = {
    0: "Enjoyment",
    1: "Sadness",
    2: "Anger",
    3: "Fear",
    4: "Disgust",
    5: "Surprise",
    6: "Other"
}

predicted_emotion = emotion_map[predicted_class]
print(f"Text: {text}")
print(f"Predicted emotion: {predicted_emotion}")

License

This model is released under the Apache-2.0 license.

Acknowledgments

• Base model: vinai/phobert-base-v2
• Dataset: VSMEC (Vietnamese Social Media Emotion Corpus)
• ViSoLex Toolkit