SentimentBot: A Sentiment Analysis Chatbot with BERT 🤖

Overview

SentimentBot is a fine-tuned BERT model (bert-base-uncased) designed for sentiment analysis, trained on the SST-2 (Stanford Sentiment Treebank) dataset. This model predicts whether a given text expresses a positive or negative sentiment with high accuracy, achieving 92.3% accuracy and 92.3% F1 score on the validation set after 2 epochs. It was developed as part of an AI Engineer internship project focusing on NLP.

Model Details

• Base Model: bert-base-uncased
• Dataset: SST-2 (2 classes: 0 = negative, 1 = positive)
• Training: Fine-tuned for 2 epochs with a batch size of 16
• Performance: Eval Loss: 0.311 Accuracy: 92.3% F1 Score: 92.3%

Usage

To use the SentimentBot model in your Python project, install the required libraries and load the model as follows:

pip install transformers torch

from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch

# Load the model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("moraix/SentimentBot")
model = AutoModelForSequenceClassification.from_pretrained("moraix/SentimentBot").to("cuda")

# Example input
text = "I love this movie so much!"
inputs = tokenizer(text, return_tensors="pt", padding="max_length", truncation=True, max_length=128).to("cuda")

# Predict sentiment
model.eval()
with torch.no_grad():
    outputs = model(**inputs)
    predictions = torch.softmax(outputs.logits, dim=1)
    predicted_class = torch.argmax(predictions, dim=1).item()
    confidence = predictions[0, predicted_class].item()

label_map = {0: "negative", 1: "positive"}
sentiment = label_map[predicted_class]
print(f"Text: {text}")
print(f"Sentiment: {sentiment} (Confidence: {confidence:.2f})")

Intended Uses

• Primary Use: Analyzing sentiment in English text (e.g., movie reviews, social media posts).
• Out-of-Scope: Multi-class sentiment (e.g., neutral) or non-English text (requires further fine-tuning).

Limitations

• Trained only on SST-2, so performance may vary on other datasets.
• Limited to binary classification (positive/negative).

Future Improvements

• dd support for more complex sentiments (e.g., neutral, angry).
• Expand the dataset with custom data.
• Deploy the chatbot as a web app using Flask or Streamlit.

Contributing

Feel free to fork this repository, open issues, or submit pull requests. Feedback is always welcome! Contact For questions, reach out via mohammadmehdiomidi95@gmail.com or connect with me on LinkedIn.