Model Card for uvegesistvan/wildmann_german_proposal_2b

Model Overview

This model is a multi-class emotion classifier trained to identify nine distinct emotional states in text. The classes and their corresponding labels are as follows:

• Class 0: Anger
• Class 1: Fear
• Class 2: Disgust
• Class 3: Sadness
• Class 4: Joy
• Class 5: Enthusiasm
• Class 6: Hope
• Class 7: Pride
• Class 8: No emotion

Dataset and Preprocessing

The dataset combines original and synthetic data to improve class balance and performance. Below are the evaluation metrics for the model:

Class	Precision	Recall	F1-Score	Support
Anger (0)	0.54	0.60	0.57	777
Fear (1)	0.85	0.76	0.80	776
Disgust (2)	0.91	0.95	0.93	776
Sadness (3)	0.87	0.84	0.86	775
Joy (4)	0.84	0.81	0.83	777
Enthusiasm (5)	0.64	0.62	0.63	776
Hope (6)	0.53	0.55	0.54	777
Pride (7)	0.75	0.81	0.78	776
No emotion (8)	0.67	0.65	0.66	1553

Overall Metrics

• Accuracy: 0.72
• Macro Average: Precision = 0.73, Recall = 0.73, F1-Score = 0.73
• Weighted Average: Precision = 0.73, Recall = 0.72, F1-Score = 0.73

Performance Insights

The model achieves strong performance across most classes, particularly for "Sadness" and "Disgust." However, "Enthusiasm" and "Hope" exhibit lower recall and precision, suggesting potential areas for improvement. Future development could include targeted data augmentation or specialized techniques to handle these classes.

Model Usage

Applications

• Emotion classification in text-based datasets.
• Analyzing emotional tone in social media, reviews, or other text corpora.
• Understanding emotional context for human-computer interaction.

Limitations

• Performance varies across classes, with some (e.g., "Enthusiasm" and "Hope") showing lower metrics.
• The model may not generalize well to domains outside the training data.
• Ambiguities in text can lead to misclassification, especially for overlapping emotional states.

Ethical Considerations

The model's predictions might not always align with human interpretations of emotions, particularly in ambiguous or context-dependent cases. Misclassification could lead to inappropriate conclusions if used in sensitive applications (e.g., mental health monitoring or automated decision-making).

Future Work

• Improving performance on underrepresented classes using advanced augmentation or transfer learning techniques.
• Exploring the model's performance in multi-domain datasets.
• Adding explainability features to enhance trustworthiness in sensitive applications.