DistilBERT Goodreads Genre Classifier
A fine-tuned DistilBERT (distilbert-base-cased) model for classifying Goodreads book reviews into 8 genres.
Model Details
- Developed by: Prakhar
- Model type: Transformer (DistilBERT) for Sequence Classification
- Language: English
- Finetuned from:
distilbert-base-cased
- Number of labels: 8
Genres (Labels)
| ID |
Genre |
| 0 |
children |
| 1 |
comics_graphic |
| 2 |
fantasy_paranormal |
| 3 |
history_biography |
| 4 |
mystery_thriller_crime |
| 5 |
poetry |
| 6 |
romance |
| 7 |
young_adult |
Training Details
Training Data
- Source: UCSD Goodreads Book Graph
- Size: 6,400 training samples (800 per genre), 1,600 test samples (200 per genre)
- Preprocessing: Tokenized with DistilBertTokenizerFast, max_length=512
Training Hyperparameters
- Epochs: 3
- Batch size: 2 (effective 10 with gradient accumulation of 5)
- Learning rate: 5e-5
- Weight decay: 0.01
- Warmup steps: 100
- Training regime: fp16 mixed precision
- Training time: ~18 minutes on NVIDIA GPU
Evaluation Results
| Metric |
Score |
| Accuracy |
0.6081 |
| Weighted F1 |
0.6054 |
| Weighted Precision |
0.6034 |
| Loss |
1.2691 |
Per-Genre Performance
| Genre |
Precision |
Recall |
F1-Score |
| children |
0.62 |
0.66 |
0.64 |
| comics_graphic |
0.76 |
0.77 |
0.77 |
| fantasy_paranormal |
0.43 |
0.43 |
0.43 |
| history_biography |
0.59 |
0.60 |
0.60 |
| mystery_thriller_crime |
0.60 |
0.62 |
0.61 |
| poetry |
0.78 |
0.81 |
0.79 |
| romance |
0.62 |
0.61 |
0.61 |
| young_adult |
0.43 |
0.36 |
0.39 |
How to Use
from transformers import pipeline
classifier = pipeline("text-classification", model="Prakhar54-byte/distilbert-goodreads-genre-classifier")
result = classifier("A thrilling mystery novel with unexpected twists")
print(result)
Environmental Impact
- Hardware Type: NVIDIA GPU (3.68 GiB VRAM)
- Hours used: ~0.3 hours
- Carbon Emitted: Minimal (local training)
