Movie-recommender

About The Project

This repository contains the source code for the movie recommender app which can be found at

http://www.movierecommender.org

The app uses a matrix factorization model to learn user and movie embeddings. These embeddings can then be used to find similar movies or to recommend movies to a new user.

Features

Currently two features are supported: movie recommendation based on your IMDb ratings and similar movie search:

Movie recommendations are obtained by using your ratings (and corresponding movie embeddings) to train your own user embedding. This user embedding is then used to compute the full movie-user matrix. From this matrix, the (predicted) highest-rated movies are found.
In similar movie search you enter one or more movie titles. The embeddings of all the entered movies are averaged out and you are shown other movies whose embeddings most closely match the averaged embedding obtained earlier.

Getting Started

This section contains the steps required to reproduce the app locally. Depending on how closely you want to replicate the app, some steps can possibly be ignored.

Prerequisites

docker engine and docker-compose
postgres service
python-dotenv (pip3 install python-dotenv==0.15.0)
yoyo_migrations (pip3 install yoyo_migrations==7.2.0)

Repository setup

Clone the repository
Obtain a Movie API key at http://www.omdbapi.com (this project uses the poster API)
Add the API key to the .env.example file and rename the file to .env

Encoding setup

I do not provide the user/movie encodings - you will have to train those by yourself. There are many freely available datasets which can be used.

Database setup

You will need to generate a database dump with the movie and user encodings, titles and IMDb ids to be used by the app. This database dump will be used by docker during containerization to initialize the (docker) postgres database.

Make sure the postgres service is running and that a database matching the name in the .env file exists.
Run python3 migrate.py

You should now have a database with multiple empty tables. It is up to you to fill the movies and users tables with the encodings you have trained. You should also fill the movies table with the IMDb ids and titles of the movies from your training dataset.

Once the tables have been filled, generate a database dump with, e.g.

sudo -H -u postgres pg_dump db > db.sql

Running

Execute docker-compose up
Once the containers are running, go to 0.0.0.0:5000

Asa-Nisi-Masa/movie-recommender