Contextual Chatbot

A Python-based contextual chatbot that uses Natural Language Processing (NLP) and a neural network to understand and respond to user queries. The chatbot is trained on custom intents and can provide contextual responses based on user input.

Features

• Natural Language Understanding using NLTK
• Neural Network-based response generation
• Customizable intents and responses
• Context-aware conversation handling
• Probability-based response selection
• GPU support for faster training (if available)

Prerequisites

• Python 3.x
• PyTorch
• NLTK
• NumPy

Installation

1. Clone the repository:

   git clone https://github.com/remicku/contextual-chatbot.git
   cd contextual-chatbot

2. (Optional) Create the conda environment if it doesn't exist:

   conda create -n pytorch python=3.9

3. Activate the environment:

   conda activate pytorch

4. Install the required packages:

   pip install torch nltk numpy

5. Download required NLTK data:

   import nltk
   nltk.download('punkt')

Project Structure

• chat.py - Main chatbot interface (interactive CLI)
• train.py - Script for training the neural network model
• model.py - Neural network model architecture
• nltk_utils.py - NLP utility functions (tokenization, BoW)
• intents.json - Training data containing patterns and responses
• data.pth - Saved model state and metadata

Usage

1. Train the model (creates data.pth):

   python  train.py

2. Start chatting:

   python  chat.py

3. Deactivate the environment when you're done:

   conda deactivate

Type 'quit' to exit the chat.

Customization

You can customize the chatbot's responses by modifying the intents.json file. The file structure should be:

{
  "intents": [
	{
	  "tag": "greeting",
	  "patterns": ["Hi", "Hello", "Hey"],
	  "responses": ["Hello!", "Hi there!", "Hey!"]
	}
    // Add more intents as needed
  ]
}

Changes to this file require retraining the model (python train.py).

How It Works

1. Input Preprocessing: User input is tokenized and stemmed using NLTK
2. Vectorization: The text is converted to a bag-of-words vector
3. Prediction: A neural network processes the input and predicts the most likely intent (responses are selected based on the predicted intent and confidence threshold)
4. Confidence Check: If the model's confidence > 75%, a matching response is chosen
5. Fallback: Otherwise, a default fallback message is returned

Model Architecture

The neural network consists of:

• Input layer: Size = vocabulary length
• Two hidden layers: 8 neurons each, ReLU activation
• Output layer: Size = number of unique intents
• Loss function: CrossEntropyLoss
• Optimizer: Adam

Contributing

Contributions are welcome! Feel free to submit issues or open pull requests to improve this chatbot.