BERT IMDb Sentiment Classifier

A fine-tuned BERT-base-uncased model for binary sentiment classification of IMDb movie reviews.

The model was fine-tuned using PyTorch and the Hugging Face Transformers Trainer API.

Model Details

• Base model: bert-base-uncased
• Model type: BertForSequenceClassification
• Task: Text Classification / Sentiment Analysis
• Language: English
• Dataset: IMDb Movie Reviews
• Number of labels: 2
• Max sequence length: 512

Labels

0 → negative
1 → positive

Intended Uses

Direct Use

This model can be used directly for binary sentiment classification of English movie reviews (or similar review-style text):

• negative — negative sentiment
• positive — positive sentiment

Out-of-Scope Use

• Non-English text
• Multi-class or fine-grained sentiment / emotion detection
• Long documents significantly exceeding 512 tokens (truncation will occur)
• Domains very different from movie reviews without further evaluation

How to Use

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

model_id = "cannizaroo/imdb-bert-sentiment"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForSequenceClassification.from_pretrained(model_id)
model.eval()

text = "This movie was absolutely fantastic. I loved every minute."

inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True, max_length=512)

with torch.no_grad():
    outputs = model(**inputs)
    predicted_id = outputs.logits.argmax(dim=-1).item()

print(model.config.id2label[predicted_id])  # "positive" or "negative"

Using the pipeline API:

from transformers import pipeline

classifier = pipeline("text-classification", model="cannizaroo/imdb-bert-sentiment")
classifier("This movie was absolutely fantastic. I loved every minute.")
# [{'label': 'positive', 'score': ...}]

Training Data

IMDb Movie Reviews (imdb):

• 25,000 training reviews
• 25,000 test reviews
• Binary labels: 0 → Negative, 1 → Positive
• Text tokenized with the BERT tokenizer (input_ids + attention_mask), with padding and truncation

Training Procedure

Fine-tuned with:

• PyTorch
• Hugging Face Transformers Trainer
• Hugging Face Datasets
• Hugging Face Evaluate
• Google Colab GPU

Training Hyperparameters

Learning Rate: 2e-5
Batch Size: 16
Epochs: 2
Weight Decay: 0.01
Maximum Sequence Length: 512

Evaluation

Evaluated on the final held-out IMDb test set:

Model: BERT-base-uncased + Fine-tuning

Accuracy:  93.21%
Precision: 95.53%
Recall:    90.66%
F1 Score:  93.03%

Test Set Confusion Matrix

                  Predicted
                Negative  Positive

Actual Negative   11970      530
Actual Positive    1167    11333

Metrics computed with Hugging Face Evaluate and scikit-learn.

Limitations and Bias

• Trained only on IMDb movie reviews; performance may degrade on other domains (e.g., tweets, product reviews, formal text).
• English only.
• May reflect biases present in IMDb data and in the original bert-base-uncased pretraining corpus.
• Truncates inputs longer than 512 tokens, so sentiment expressed late in very long reviews may be lost.

Citation

If you use this model, please cite the base model and dataset:

@article{devlin2018bert,
  title={BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding},
  author={Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina},
  journal={arXiv preprint arXiv:1810.04805},
  year={2018}
}

@inproceedings{maas2011learning,
  title={Learning Word Vectors for Sentiment Analysis},
  author={Maas, Andrew L. and Daly, Raymond E. and Pham, Peter T. and Huang, Dan and Ng, Andrew Y. and Potts, Christopher},
  booktitle={Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics},
  year={2011}
}