ds-claudia/classify_emotions_into_six_categories_with_distilbert

results

This model is a fine-tuned version of distilbert-base-uncased on the Emotions dataset. It achieves the following results on the evaluation set:

• Loss: 0.4624
• Accuracy: 0.84
• F1 Score: 0.8394
• Precision: 0.8399
• Recall: 0.84
• Auc Score: 0.9720

Model description

This model is a fine-tuned version of DistilBERT, a smaller and faster version of BERT, for emotion classification tasks. It was trained on the Emotions dataset, which contains text labeled with six emotion categories: joy, sadness, anger, love, fear, surprise. The model processes input text to predict the underlying emotion, making it suitable for sentiment analysis and emotion detection tasks.

Intended uses & limitations

Intended Uses:

• Detecting emotions in text for applications such as customer support, social media monitoring, or mental health analysis.
• Can be applied in any domain where understanding emotional tone is valuable, including surveys, feedback forms, and user reviews.

Limitations:

• The model may struggle with out-of-domain text or highly specialized terminology that wasn't present in the training data.
• Emotion detection in short, ambiguous, or highly context-dependent sentences may lead to misclassification.
• The model was trained on a dataset that might not fully represent all cultural or linguistic variations of emotions.

Training and evaluation data

Training Data: The model was trained on training set portion of the Emotions dataset, which contains text labeled with 6 emotion categories: "anger," "fear," "joy," "love," "sadness," and "surprise." The dataset consists of user-generated text from various online sources.

Evaluation Data: The model's evaluation was done on the test set portion of the Emotions dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1 Score	Precision	Recall	Auc Score
1.2004	0.5	500	0.7984	0.6965	0.6545	0.7094	0.6965	0.9250
0.7255	1.0	1000	0.6070	0.7895	0.7864	0.7895	0.7895	0.9565
0.6118	1.5	1500	0.5277	0.8205	0.8193	0.8210	0.8205	0.9650
0.5542	2.0	2000	0.4941	0.828	0.8268	0.8290	0.828	0.9689
0.5113	2.5	2500	0.4688	0.837	0.8358	0.8365	0.837	0.9714
0.506	3.0	3000	0.4624	0.84	0.8394	0.8399	0.84	0.9720

Framework versions

• Transformers 4.48.3
• Pytorch 2.6.0
• Datasets 3.2.0
• Tokenizers 0.21.0