distilbert-sentiment-5class

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

checkpoints

This model is a fine-tuned version of distilbert-base-uncased trained on processed Amazon review datasets from multiple categories. It is designed for sentiment classification across five classes. It achieves the following results on the evaluation set:

• Loss: 0.9056
• Accuracy: 0.6087
• F1 Macro: 0.6101
• F1 Weighted: 0.6085
• Precision Class 1: 0.6762
• Precision Class 2: 0.4900
• Precision Class 3: 0.5177
• Precision Class 4: 0.6173
• Precision Class 5: 0.7535
• Recall Class 1: 0.6781
• Recall Class 2: 0.5293
• Recall Class 3: 0.5049
• Recall Class 4: 0.5606
• Recall Class 5: 0.7781
• F1 Class 1: 0.6772
• F1 Class 2: 0.5089
• F1 Class 3: 0.5112
• F1 Class 4: 0.5876
• F1 Class 5: 0.7656

Model description

This model is a distilled, lightweight transformer optimized for multi-class sentiment analysis. It is capable of processing text efficiently while maintaining competitive performance. The model was fine-tuned on Amazon product reviews spanning different categories such as electronics and home goods... It can classify reviews into five sentiment levels, from highly negative to highly positive.

Intended uses & limitations

Automated sentiment classification for Amazon reviews or similar e-commerce datasets.

Analysis of customer feedback trends to inform product decisions.

Training and evaluation data

The model was trained on a processed subset of Amazon review datasets covering multiple product categories. Text data was cleaned, tokenized, and labeled into five sentiment classes. Evaluation was performed on a held-out test set representing the same categories to ensure balanced coverage.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1