Analytics on Amazon Reviews

Data Analytics exam final project, MSc in Computer Science.

By Matteo Coppola, Luca Palazzi, Antonio Vivace.

Exploration, Sentiment Analysis, Topic Analysis (LDA) and a VueJS web application exposing the trained models.

GO. PLAY. WITH THE PLOTS. (web demo deployment)

Documentation

Exploration

Web demo

Run

Set up the a Python virtual environment and install required packages

cd scripts
python3 -m venv .
source bin/activate
pip3 install -r requirements.txt
python3 -m spacy download en

Optionally, install a ipynb kernel to use the venv packages

pip3 install --user ipykernel
python -m ipykernel install --user --name=myenv
# Check the installed kernels
jupyter kernelspec list
# Run Jupyter
jupyter lab

Now, to run the full pipeline:

python3 main.py

A Flask application exposes a simple API (on port 5000) allowing the trained models to be used on demand via simple HTTP requests (in main.py). The VueJS application needs a recent version of NodeJS and npm.

cd webapp
npm install
# serve the web application with hot reload at localhost:8080/reviews-sentiment
npm run serve
# builds the web application for production
npm run build
# deploys the build on the master branch, making github serve it on https://avivace.github.io/reviews-sentiment
npm run deploy

Antuz notes

Accent is #B71C1C, typeface is Barlow 500. On the plots and graphs, typeface is Inter 600, palette is #4DAF4A, #FF7F00, #C73E31.

Final notes from our supervisor, E.Fersini

Unverified/Spam "boom" happens around the first-publishing of some product, aggregating data from a category will hardly show this (there are papers on this)