sentiment-analysis-en-v1

Multimodal Emotion Analyzer for 26 emotions

Model Description

This is a multimodal emotion analysis model that processes both audio and text inputs to predict 26 different emotions with regression scores (0-1).

Architecture

• Text Encoder: RoBERTa-base
• Audio Encoder: CNN-based encoder with 3 convolutional layers
• Fusion: MLP layers with ReLU activation and Dropout
• Output: 26 emotion regression scores (0-1 range)

Emotion Labels

admiration, aesthetic appreciation, awe, anxiety, fear, horror, disgust, calmness, romantic love, sexual desire, nostalgia, interest, surprise, excitement, anger, pride, triumph, contempt, disappointment, empathic pain, sadness, guilt, envy, amusement, awkwardness, adoration

Model Files

• best_model.pth: Trained model weights
• config.json: Model configuration
• training_history.png: Training progress visualization

Usage Example

import torch
from transformers import AutoTokenizer
from models.multimodal_emotion_all import MultimodalEmotionAnalyzerAll

# Load model
model = MultimodalEmotionAnalyzerAll(num_emotions=26)
checkpoint = torch.load('best_model.pth', map_location='cpu')
model.load_state_dict(checkpoint['model_state_dict'])
model.eval()

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained("roberta-base")

# Process text
text = "I am feeling happy today"
inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)

# Get predictions
with torch.no_grad():
    # Note: This model expects both text and audio inputs
    # For text-only inference, you may need to modify the forward pass
    outputs = model(**inputs)
    emotion_scores = torch.sigmoid(outputs)

Training Details

• Training Date: 2025-08-26
• Number of Emotions: 26
• Model Type: Multimodal (Text + Audio)
• Base Model: RoBERTa-base
• Loss Function: MSE Loss
• Optimizer: AdamW

Performance

• Model Size: 1432.7 MB
• Parameters: 125,172,565

Citation

If you use this model in your research, please cite:

@misc{emotion_analyzer_2024, title={Multimodal Emotion Analyzer}, author={Your Name}, year={2024}, url={https://huggingface.co/CURI-AI/sentiment-analysis-en-v1} }