wncelrcn/mindmap-deBERTa-goemotions-multilabel

🧠 MindMap Emotion Classifier (deBERTa v3 + GoEmotions)

A fine-tuned multi-label emotion classification model based on microsoft/deberta-v3-base and trained on the GoEmotions dataset. This model is designed to power emotional tagging for personal journaling and mental wellness applications like MindMap.

Note: This model is part of the experiment to find the best-performing emotion classification model for our digital journaling web application called MindMap.

🚀 Model Details

• Base Model: DeBERTa v3 Base
• Task: Multi-label emotion classification
• Dataset: GoEmotions (27 emotions + neutral)
• Activation: Sigmoid
• Loss: BCEWithLogitsLoss
• Output: Probability scores for each of the 28 emotion labels

🏷️ Supported Emotions (28 classes):

admiration, amusement, anger, annoyance, approval, caring, confusion, curiosity, desire, disappointment, disapproval, disgust, embarrassment, excitement, fear, gratitude, grief, joy, love, nervousness, optimism, pride, realization, relief, remorse, sadness, surprise

📥 How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_name = "wncelrcn/mindmap-deBERTA-goemotions-multilabel"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
model.eval()

# Sample journal entry
text = "I felt overwhelmed today. I didn't want to get out of bed. But a short walk helped me clear my mind."

# Tokenize input
inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)

# Get prediction
with torch.no_grad():
    outputs = model(**inputs)
    logits = outputs.logits
    probs = torch.sigmoid(logits)[0]  # Multi-label output

# Map to labels
labels = model.config.id2label
predictions = [(labels[i], float(prob)) for i, prob in enumerate(probs) if prob > 0.05]

# Display
for label, score in sorted(predictions, key=lambda x: -x[1]):
    print(f"{label}: {score:.3f}")

📊 Evaluation

This model was evaluated using:

Metrics: F1-score (micro/macro), Precision, Recall

Validation split: 90/10 on the simplified GoEmotions dataset

Threshold: 0.05 for emotion label activation

🧠 Use Case

Originally fine-tuned for MindMap, a digital journaling app that helps users track and reflect on their emotional well-being. The model enables emotion-aware feedback and visualizations, offering therapeutic insight to users based on their writing.

📦 Model Files

pytorch_model.bin: Model weights

config.json: Model configuration

tokenizer.json, tokenizer_config.json: Tokenizer details

special_tokens_map.json, vocab.json: Tokenizer vocabulary

📚 Citation / Credit

Base model: Microsoft DeBERTa v3

Dataset: GoEmotions by Google Research

🛠 Maintained by @wncelrcn