This program is written in Python and employs several critical libraries and packages, including:The following is an implementation of a Natural Language Processor in the form of a ChatBot which takes in a user's question or comment, and, implementing common Natural Language Processor techniques as well as probability, returns an answer
- Pytorch
- An open sourced Machine Learning library used for natural language processing
- N.L.T.K (Natural Language Toolkit)
- A bundle of libraries for English language processing
Tokenizing
- The first function in preparing our user input for interpretability is a Tokenizer, which breaks a sentence down into it's individual words
Stemming
- Next, we apply a stemming function, which further breaks down each word into it's root without affixes or suffixes
This allows a better, more consistent interpretation
Bag-Of-Words
- Further divide's our stemmed and tokenized words into occurences within a given text. This allows a computer to interpret information processed as language into fathomable, quantifiable vectors and numbers
Specifically, for each occurence of a word, a value of 1 is given at the index where the word occurs, otherwise a value of 0 is given
After our text is broken down into it's most basic, interpretable form, we are able to train our model on the text, and apply a probability metric to display specific responses
Model
- Our model is a simple neural network consistent of two linear "feed forward" layers
python3 train.py
This will deploy all required training
python3 chat.py
This will deploy the chatbot in the terminal with realistic responses within any given context!
In the future, we will implement a greater amount of hidden layers, as well as an increased amount of text located within the .json file on which to train and broaden the capability of the chatbot