Seq2Seq-Chatbot

Introduction

This Chatbot is a TensorFlow implementation of Seq2Seq Mode. It make use of a seq2seq model RNN for sentence predictions. The chatbot is trained on Cornell Movie Dialogs Corpus on Conversation of lenghth between 5 to 2 words

Publication

With the given link below,you can find my paper on Seq2Seq AI Chatbot with Attention Mechanism

Link: https://arxiv.org/abs/2006.02767

arix identifier: 2006.02767 License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Installation

To run this chabtot you will need:

• Python 3.5
• Tensorflow (1.14.0)
• Numpy
• Flask
• Matplotlib
• Dataset

Running

To chat with the trained model from Web Chat:

1. First install the required depedencies and run pip install -r requirment.txt
2. From the BOT_UI directory run run.py python run.py
3. Open a browser to the URL indicated by the server console, This is typically: http://localhost:5000/

To chat with the trained model from Python Console:

1. First install the required depedencies from root directory and run pip install -r requirment.txt
2. From the root directory run python chat.py

List of Hyperparameters

For Training Model

To Train Model from Python Console:

1. Configure the config.py file for desired training hyperparameters
2. Run train.py python train.py
3. Make sure the datasets should be present in the datasets folder

Results

Config1:Model was able to train to Accuracy of 62% and Loss of 19% and after that it was overfitting Config2:Model was able to train to Accuracy of 63% and Loss of 18% and after that it was overfitting

Accuracy

Config1

Loss

Config2

Accuracy

Loss

Chat Results

Human :hey
Bot :hi
Human :what doing
Bot :sleep
Human :when will you be okay
Bot :just few days 
Human :where are you from
Bot :southern california
Human :who am i
Bot :doctor livingston
Human :really well
Bot :goodbye

Bot reply Result in web app