Project Summary
This project demonstrates the development of an AI-powered chatbot with intent classification capabilities, comparing the performance of Naive Bayes and BERT models. The chatbot integrates the best-performing model to classify user intents and provide meaningful responses.
Key Features
- Naive Bayes Model: Utilizes TF-IDF vectorization for feature extraction. Efficient for resource-constrained settings.
- BERT Model: Fine-tuned transformer with superior contextual understanding and high accuracy.
- Chatbot Integration: Flask-based REST API for interaction using the trained BERT model.
Evaluation Results
| Model | Accuracy | Precision | Recall | F1-Score |
|---|---|---|---|---|
| Naive Bayes | 79% | 78% | 79% | 78% |
| BERT | 89% | 88% | 89% | 88% |
Observations
- Naive Bayes: Lightweight and efficient for resource-constrained settings.
- BERT: Superior performance with better contextual understanding, ideal for production.
Recommendations
- High accuracy: Use BERT for nuanced intent classification.
- Efficiency: Leverage Naive Bayes for lightweight scenarios.
- Expand capabilities: Add more intents and responses to enhance the chatbot's functionality.