Multilingual Chatbot

Course Project - CS 521 Statistical NLP - Spring 2021

Team Members: Meghana Jagadish and Samujjwaal Dey

Overview

A chatbot for Pizza ordering service, as per user customization, supporting queries and the responses in more than one language.

Conventional Chatbots though are very useful, lack support for local languages. The goal of this project is to overcome that disadvantage and create a multilingual chatbot supporting at least 2-3 languages.

Project Tree

.
├── Readme.md
├── Results.pdf
├── chatgui.py
├── models
│   ├── chatbot_model.h5
│   └── lid.176.ftz
├── requirements.txt
├── resources
│   ├── classes.pkl
│   ├── intents.json
│   ├── pizza.png
│   └── vocab.pkl
├── train_chatbot.py
└── utils
    ├── api_utils.py
    ├── chatbot_utils.py
    ├── fasttext_utils.py
    ├── transformer_utils.py
    └── translation_utils.py

Application Description

File	Description
intents.json	Domain specific intents dataset for chatbot
vocab.pkl	Serialized chatbot model vocabulary
classes.pkl	Serialized chatbot intent classes
train_chatbot.py	Script to train the chatbot from intents
chatbot_model.h5	Locally saved, trained chatbot model
chatgui.py	Script to create chatbot GUI
chatbot_utils.py	Define functions for using chatbot model
translation_utils.py	Define functions for text translation
fasttext_utils.py	Define functions for fasttext language detection
lid.176.ftz	Locally saved fasttext pre-trained model
transformer_utils.py	Define functions for loading transformer models
api_utils.py	Define functions to use Hugging Face Inference API

Instructions to Execute

Clone this repository from GitHub and open the root directory in the terminal.

Install Requirements

The required pip packages for successfully executing the project are:

numpy==1.20.1
wget==3.2
requests==2.25.1
googletrans==3.1.0a0
fasttext==0.9.2
nltk==3.6.1
Keras==2.4.3
transformers==4.5.1
Pillow==8.2.0

Train Chatbot Model

python train_chatbot.py 

The intents are parsed from intents.json to generate the chatbot vocabulary and train a Keras sequential model for 200 epochs.

Run Chatbot GUI

python chatgui.py

The GUI window for the chatbot is created.

Each user input and chatbot response (with source language and translation) are printed in the terminal too.

Results

Results for execution of the Multilingual Chatbot for Pizza Ordering can be found in this file.

The results file shows chatbot conversations for 4 case scenarios along with screenshot of GUI.