Model Card for uvegesistvan/wildmann_german_proposal_2a

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:

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

Dataset and Preprocessing

The dataset combines original and synthetic data to improve class balance and performance. Synthetic data augmentation was applied to classes with lower representation in the original dataset, specifically "Fear," "Disgust," "Sadness," "Joy," and "Pride." The following table summarizes the distribution of original and synthetic data across training, testing, and validation sets:

Training Data:

Label	Original Count	Original (%)	Synthetic Count	Synthetic (%)
Anger	6210	100.00	0	0.00
Fear	2534	40.81	3676	59.19
Disgust	845	13.60	5366	86.40
Sadness	2670	42.99	3541	57.01
Joy	3420	55.07	2790	44.93
Enthusiasm	4347	70.00	1863	30.00
Hope	6210	100.00	0	0.00
Pride	2834	45.63	3377	54.37
No emotion	6210	100.00	0	0.00

Testing Data:

Label	Original Count	Original (%)	Synthetic Count	Synthetic (%)
Anger	777	100.00	0	0.00
Fear	317	40.85	459	59.15
Disgust	106	13.66	670	86.34
Sadness	333	42.97	442	57.03
Joy	428	55.08	349	44.92
Enthusiasm	543	69.97	233	30.03
Hope	777	100.00	0	0.00
Pride	354	45.62	422	54.38
No emotion	777	100.00	0	0.00

Validation Data:

Label	Original Count	Original (%)	Synthetic Count	Synthetic (%)
Anger	776	100.00	0	0.00
Fear	317	40.80	460	59.20
Disgust	105	13.53	671	86.47
Sadness	334	42.99	443	57.01
Joy	427	55.03	349	44.97
Enthusiasm	544	70.01	233	29.99
Hope	776	100.00	0	0.00
Pride	354	45.62	422	54.38
No emotion	776	100.00	0	0.00

Evaluation Metrics

The model was evaluated using precision, recall, F1-score, and support for each class. Below are the detailed metrics:

Class	Precision	Recall	F1-Score	Support
Anger (0)	0.57	0.64	0.61	777
Fear (1)	0.84	0.77	0.80	776
Disgust (2)	0.91	0.95	0.93	776
Sadness (3)	0.84	0.85	0.85	775
Joy (4)	0.78	0.85	0.81	777
Enthusiasm (5)	0.63	0.63	0.63	777
Hope (6)	0.51	0.55	0.53	777
Pride (7)	0.77	0.77	0.77	776
No emotion (8)	0.47	0.34	0.39	777

Overall Metrics

• Accuracy: 0.71
• Macro Average: Precision = 0.70, Recall = 0.71, F1-Score = 0.70
• Weighted Average: Precision = 0.70, Recall = 0.71, F1-Score = 0.70

Performance Insights

The model achieves strong performance across most classes, particularly for "Disgust" and "Sadness." However, the "No emotion" class shows lower recall, which could indicate challenges in distinguishing neutral text from emotional expressions. Additional fine-tuning or data augmentation may help address this limitation.

Model Usage

Applications

• Emotion classification in text-based datasets.
• Analyzing emotional tone in social media, reviews, or other text corpora.

Limitations

• Performance varies across classes, with some (e.g., "Hope" and "No emotion") showing lower recall.
• The model may not generalize well to domains outside the training data.

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).