h2o2.ai presents: Voice of H2O

A production-ready web application for Voice of Customer (VOC) analysis

From retail to big tech, customers can be hard to read - and that's where we come in. Introducing "Voice of H2O", our cutting-edge web application for Voice of Customer analysis. Built on H2OWave, our intuitive interface and advanced analytical capabilities allow firms to bridge the gap between big data and individual customers.

Our application allows you to:

1. Explore key themes in customer feedback
2. Visualise sentiments of customers
3. Perform trend analysis over time

Whether you're a small business owner or a large enterprise, our VoC analysis application will help you improve customer satisfaction, boost loyalty, and drive business growth. And better yet - you can do this all for free. Simply follow setup instructions to containerise and build our web application on your very own device.

Setup Instructions

Option 1: Docker

Deployment

To build Docker image and container, run the following commands on h2o2.ai project folder:

docker compose up

Running Pipeline via Notebook

Note that you will need to have the necessary computing power to run model training on Docker. Else the terminal throws an error showing DeadKernelError: Kernel died.

To run all pipelines via notebook, run the following commands on final_presentation folder:

python3 run_notebook.py

Running Pipeline via Terminal

Note that you will need to have the necessary computing power to run model training on Docker. Else running the command to train sentiment analysis models in the terminal might not run through the epochs and kills the training (showing 'Killed').

To run all pipelines via terminal, run the following commands on h2o2.ai project folder:

# Preprocess training data. This step should be done before calling train functions.
python3 -m src.preprocessing.transformations

# Train sentiment analysis models
python3 -m src.models.sentiment_analysis.train.train

# Train topic modelling models
python3 -m src.models.topic_modelling.train.train

# Predict test data. Test file path should be supplied from config file.
python3 -m src.models.predict

# Execute unit testing.
python3 -m src.unittest.unit_testing

Running App

Open the app on http://127.0.0.1:8080/ or http://localhost:8080/

Option 2: EC2 Instance

Unfortunately, it's not possible to run H2O Wave app on EC2 Instance given the limited access to AWS provided on RLCatalyst as it needs further setup on AWS to open the relative port of EC2 Instance (https://h2o.ai/blog/deploy-a-wave-app-on-an-aws-ec2-instance/).

Hence, EC2 Instance is only used for model training and test prediction purposes.

Prerequisites

You will need to have an ubuntu EC2 GPU instance running and connect to the instance via terminal.

Then, copy all files to the instance.

Environment Setup

To setup the EC2 instance, run the following commands on h2o2.ai project folder:

# Install pip for Python 3
sudo apt install python3-pip

# Install all libraries needed
pip install -r requirements.txt

Running Pipeline via Notebook

To run all pipelines via notebook, run the following commands on final_presentation folder:

python3 run_notebook.py

Running Pipeline via Terminal

To run all pipelines via terminal, run the following commands on h2o2.ai project folder:

# Preprocess training data. This step should be done before calling train functions.
python3 -m src.preprocessing.transformations

# Train sentiment analysis models
python3 -m src.models.sentiment_analysis.train.train

# Train topic modelling models
python3 -m src.models.topic_modelling.train.train

# Predict test data. Test file path should be supplied from config file.
python3 -m src.models.predict

# Execute unit testing.
python3 -m src.unittest.unit_testing

Option 3: Local

Prerequisites

You will need to have a valid Python and Conda installation on your system.

Environment Setup

To avoid library dependency issues, run the following commands on h2o2.ai project folder:

# Create environment from file
conda env create -f environment.yml

# Activate created conda environment
conda activate voc_env

Running Pipeline via Notebook

Note that you will need to have the necessary computing power to run model training on local. Else this might slow down your laptop or use up extensive resources.

To run all pipelines via notebook, run the following commands on final_presentation folder:

python run_notebook.py

Running Pipeline via Terminal

Note that you will need to have the necessary computing power to run model training on local. Else this might slow down your laptop or use up extensive resources to train the sentiment analysis model.

To run all pipelines via terminal, run the following commands on h2o2.ai project folder:

# Preprocess training data. This step should be done before calling train functions.
python -m src.preprocessing.transformations

# Train sentiment analysis models
python -m src.models.sentiment_analysis.train.train

# Train topic modelling models
python -m src.models.topic_modelling.train.train

# Predict test data. Test file path should be supplied from config file.
python -m src.models.predict

# Execute unit testing.
python -m src.unittest.unit_testing

Running App

To start app, run the following commands on h2o2.ai project folder:

wave run src.app.app

Open the app on http://localhost:10101/

Sphinx Documentation

To view the latest version of the documentation, please refer to https://shreyasriram4.github.io/h2o2.ai/

By default we have included in the github workflow to regenerate the documentation each time there is a push, pull request or workflow dispatch. However if you would like to regenerate the documentation locally, use the following command at the h2o2.ai project folder directory:

sphinx-build -b html docs docs\_build\html

The above command tells sphinx to build the documentation using the files in docs folder and outputs the html files at docs_build\html folder. Note that you have to install all the required python libraries in requirements.txt.

Thereafter, open up the index.html within the docs_build\html folder to view the documentation.