Deep Learning Based Chat Bot

Developing an intelligent chatbot using deep learning with Keras.

Steps to Follow

Establish the Required Packages

• see requirements.txt file
• install the dependencies
  • run pip install -r requirements.txt

Define Intents

• these intents are created to to establish context during a conversation
• create intents.json file

Data Prep

• these succeeding steps will be completed in the train.ipynb file -> everything is successfully ran in the train-kg.ipynb file
  • import all the required packages
  • load the json file and extract necessary data
    • the training_sentences variable is what holds the training data -> the sample messages is the intent categories -the training_labels variable holds the target labels that correspond with the training data
  • create the "LabelEncoder()" function to convert the target labels into an understandable model form
  • create tokenizer class to vectorize the data corpus
    • using this class for the pre-processing tasks removes the punctuations and splits the words into lists of tokens
    • the tokens are then indexed or vectorized
    • the oov_token attribute is used to deal with out of vocabulary tokens
  • the padded_sequences method is to make the training text sequences the same size

Model Training

• used Sequential class from Keras to define the Neural Network architecture for the proposed model
• used the fit method to train the model
• saved the required files to use at inference time

Inference

• implementing a chat function to communicate with users
• How This Works?
  • upon receiving a new message, the chat bot calculates the similarity between the new text sequence and the training data
  • with the confidence scores from each category, the model categorizes the user's message to an intent with the correlating highest confidence score
• this is all laid out in the chat.py file