GaneshArihanth/SentimentAnalysis-Bert

BERT Sentiment Analysis on Amazon Polarity:

This model fine-tunes the bert-base-uncased model from Google on the Amazon Polarity dataset to classify customer reviews into positive or negative sentiment.

---

Model Details:-

Model Description:

• Developed by: Ganesh Arihanth
• Finetuned from model: google-bert/bert-base-uncased
• Language(s): English
• License: MIT
• Model type: Transformer-based sentiment classifier
• Dataset used: fancyzhx/amazon_polarity
• Pipeline type: text-classification

Model Sources:

• Repository: GitHub Repo
• Dataset: Amazon Polarity
• Base model: BERT base uncased

Results:

Metric	Value
Accuracy	96.5%
F1 Score	96.36%
Precision	~96.3%
Recall	~96.4%

---

Uses:-

Direct Use:

• Binary sentiment classification (positive/negative) of product reviews
• Customer feedback analysis
• E-commerce analytics dashboards

Downstream Use:

• Chatbots or virtual assistants for feedback and understanding
• Review filtering systems
• Business intelligence tools for automated sentiment insights

Out-of-Scope Use:

• Not suitable for non-English text without re-training
• Not suitable for detecting sarcasm or highly nuanced sentiment
• Not trained for offensive content detection

---

Bias, Risks, and Limitations:-

• Trained only on Amazon product reviews; may not generalise well to other domains
• Risk of overfitting to Amazon-specific language
• May perform poorly on concise, ambiguous, or sarcastic reviews

Recommendations:-

• Evaluate on your domain-specific test set before deployment
• Retrain or fine-tune further if accuracy is low on your data

---

How to Use:-

import torch
from transformers import BertTokenizerFast, BertForSequenceClassification

# Set the device (use GPU if available)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")

# Load the saved model and tokenizer
model_path = "GaneshArihanth/bert-amazon-sentiment"  # Change to your actual HF repo path
tokenizer = BertTokenizerFast.from_pretrained(model_path)
model = BertForSequenceClassification.from_pretrained(model_path)
model.to(device)
model.eval()

# Function to predict sentiment
def predict_sentiment(text):
    inputs = tokenizer(
        text,
        padding="max_length",
        truncation=True,
        max_length=128,
        return_tensors="pt"
    )
    input_ids = inputs["input_ids"].to(device)
    attention_mask = inputs["attention_mask"].to(device)

    with torch.no_grad():
        outputs = model(input_ids, attention_mask=attention_mask)
        logits = outputs.logits
        prediction = torch.argmax(logits, dim=-1).item()

    sentiment = "Positive 😊" if prediction == 1 else "Negative 😖"
    return sentiment

# Example usage
while True:
    user_input = input("Enter a review (or 'exit' to quit): ")
    if user_input.lower() == "exit":
        break
    result = predict_sentiment(user_input)
    print(f"Predicted Sentiment: {result}")

---

Training Procedure:-

• Hardware: NVIDIA RTX 3060 (12GB VRAM)
• Batch Size: 64
• Epochs: 2
• Optimizer: AdamW
• Learning Rate: 2e-5
• Max Sequence Length: 512
• Mixed Precision: Enabled (FP16)
• Framework: PyTorch + Hugging Face Transformers

---

Technical Specifications:-

Model Architecture

• Model: BERT (base, uncased)

• Parameters: 110M

• Layers: 12

• Heads: 12

• Hidden Size: 768

• Task Objective: Binary classification using [CLS] token

Infrastructure

• Environment: Python 3.10+, PyTorch, Hugging Face Transformers

• Device: RTX 3060 with CUDA

---