Bakerly, the Deep Learning Chatbot

Introduction

This project is a contextual chatbot built using TensorFlow. The chatbot is designed to provide meaningful responses based on general input queries and those related to baking by leveraging natural language processing (NLP) techniques and machine learning.

Features

• Tokenization and stemming of input text for preprocessing.
• Classification of user intents using bag-of-words.
• Training a neural network to classify intents.
• Generating appropriate responses based on user queries.
• Flexible and extensible design to add more intents and responses.

Requirements

• Python 3.7 or higher
• Libraries:
  • TensorFlow
  • Numpy
  • NLTK

Refer to requirements.txt for installing all the dependencies.

Setup Instructions

1. Clone the Repository:

   git clone https://github.com/Subhoshri/Chatbot
   cd Chatbot

2. Install required Libraries:

   pip install Library-Name

3. Prepare the Environment:

   • Download NLTK data (if not already installed):

     import nltk
     nltk.download('punkt')

4. Update the intents file: Modify the intents.json file to add or update chatbot intents and patterns.

5. Run the Application:

   python app.py

6. Enter the URL provided after running the previous commands into your web browser.

Bakerly is now ready to chat!

How It Works

1. Data Preprocessing

• The intents.json file contains training data in patterns, and their associated tags (and contexts).
• Words in patterns are tokenized and stemmed using NLTK's Stemmer.
• A bag-of-words representation is generated for each pattern.

2. Model Training

• A feed-forward neural network (FFNN) is trained on the bag-of-words and output tags using TensorFlow.
• The output is a one-hot-encoded vector representing the possible intent classes.

3. Response Generation

• The trained model predicts the intent of user input.
• The bot selects an appropriate response from the responses list associated with the predicted intent.

Example Workflow

1. User Input: "Hello!"
2. Preprocessing:
   • Tokenize: ["hello"]
   • Stem: ["hello"]
   • Bag of Words: [1, 0, 0, 0, ...]
3. Model Prediction: Intent -> greeting
4. Bot Response: "Hi there! How can I assist you today?"

Further Scope of Update

• Adding New Intents:

  • Update the intents.json file with new intents, patterns, and responses.
  • Retrain the model.

• Improving Responses:

  • Use additional NLP techniques, such as named entity recognition (NER) or sentiment analysis, to enhance contextual responses.

Known Issues

• The chatbot may not handle out-of-scope queries effectively.
• Responses are limited to predefined data in the intents.json file.

References

• TensorFlow Documentation: https://www.tensorflow.org

---

License

This project is licensed under the MIT License.