IMDb Sentiment Analysis using DistilBERT

Project Overview

This project uses a fine-tuned DistilBERT transformer model to perform sentiment analysis on movie reviews. It classifies text into two categories: Positive and Negative.

The model is fine-tuned on the IMDb movie review dataset using PyTorch and Hugging Face Transformers.

Model Details

Training Details

• Training samples: 500
• Evaluation samples: 250
• Training epochs: 1
• Model: Fine-tuned DistilBERT
• Optimizer and other training settings: Refer to the training configuration.

Note: The reported evaluation accuracy was obtained using the sample evaluation setup above. It should not be interpreted as a full-dataset benchmark.

How to Use

Install the required libraries:

pip install torch transformers

Load the model and tokenizer:

import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification

model_name = "Swalpita123/imdb-sentiment-distilbert"

tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)

text = "This movie was absolutely fantastic!"

inputs = tokenizer(
    text,
    return_tensors="pt",
    truncation=True,
    padding=True,
    max_length=256
)

model.eval()

with torch.no_grad():
    outputs = model(**inputs)
    probabilities = torch.softmax(outputs.logits, dim=1)

predicted_id = torch.argmax(probabilities, dim=1).item()
sentiment = "POSITIVE" if predicted_id == 1 else "NEGATIVE"

print("Sentiment:", sentiment)
print("Confidence:", round(probabilities[0][predicted_id].item() * 100, 2), "%")

Example

Input: "This movie was excellent. The acting and storyline were amazing."

Expected output:

Sentiment: POSITIVE

Technologies Used

• Python
• PyTorch
• Hugging Face Transformers
• DistilBERT
• IMDb Dataset

Limitations

• The model was fine-tuned using a limited training sample.
• Predictions may not always reflect the correct sentiment, particularly for sarcasm, mixed opinions, or complex reviews.
• Confidence scores represent model probabilities, not guaranteed prediction accuracy.

Author

Swalpita Ray Nayak

Hugging Face: https://huggingface.co/Swalpita123

GitHub: https://github.com/Swalpita123/AI-Sentiment-Analysis