HovMan Chatbot

A simple intent-based chatbot built with PyTorch. This project uses a feedforward neural network for natural language understanding, trained on custom intents. It recognizes user input and provides context-aware responses in real time.

---

Features

• ✅ Neural network built using PyTorch
• ✅ Intent classification from JSON-based training data
• ✅ Bag-of-words NLP pipeline using NLTK
• ✅ Customizable intents and responses
• ✅ Confidence-based response filtering
• ✅ CLI interface for quick testing
• ✅ Trained model stored as data.pth

---

Model Architecture

Input Layer → Hidden Layer (ReLU) → Hidden Layer (ReLU) → Output Layer

• Input size: number of vocabulary tokens
• Hidden size: configurable (default = 8)
• Output size: number of intent tags
• Loss: CrossEntropyLoss
• Optimizer: Adam

---

Project Structure

├── intents.json         # Intent patterns & responses
├── nltk_utils.py        # Tokenization, stemming, bag-of-words
├── model.py             # NeuralNet class definition
├── train.py             # Training script (generates data.pth)
├── chat.py              # CLI chatbot interface
├── data.pth             # Saved model state
└── README.md

---

Installation & Setup

1. Clone the repository

git clone git@github.com:manuelhorvey/contextual_chatbot.git
cd hovman-chatbot

2. Install dependencies

pip install torch numpy nltk

3. Download NLTK tokenizer

import nltk
nltk.download('punkt')

4. Train the model

python train.py

5. Start chatting

python chat.py

---

Example Interaction

You: Hi
HovMan: Hi there, what can I do for you?

You: What do you sell?
HovMan: We sell coffee and tea

You: Do you take Paypal?
HovMan: We accept VISA, Mastercard and Paypal

You: Bye
HovMan: See you later, thanks for visiting

---

Customization

Want to add your own intents and responses?

• Edit intents.json:

{
  "tag": "your_tag",
  "patterns": ["user input variations"],
  "responses": ["your bot response"]
}

• Retrain the model:

python train.py

---

Performance Notes

This chatbot is not trained on large datasets and does not use advanced language models (like GPT or BERT). It's lightweight, fast, and great for small-scale or demo applications. Perfect for learning the fundamentals of chatbot development.

---

License

MIT License. Free for personal and commercial use. Attribution appreciated.

---