The Reviewer is a web platform for games reviews that automatically extracts informations like recommendation and helpfulness based only on its text.
If you see the files in this repository, you can see that are some interesting things happening. First, we have a file named steam_reviews.csv, it's a dataset of 10573 reviews made by real users in the Steam game store. That file includes not just the review, but recommendation status and other flags, like helpful and funny.
In a python notebook training.ipynb we can read and clean that data using the Natural Language Toolkit nltk, then produce TF-IDF vectorizers and classifiers for recommendation, helpful and funny estimates using SVM and KNN algorithms from scikit-learn.
These vectorizers and classifier were exported to joblib files, so a flask API could be implemented, offering predictions through HTTP requests. Also, that API persists new reviews made in system using pandas dataframes. Finnaly, a React-Bootstrap responsive front-end was developed to iterate with the flask API. There, a user can choose a game title, write a review, send it to the API and see its resulting predictions.
This project is encapsulated in docker containers and run with docker-compose. To run locally, one just needs to clone this repository, build the images using:
$ docker-compose build
having created the images, run the servers using:
$ docker-compose up