movierecommendations's Introduction

movieRecommendations

A simple django based movie recommendation engine. It uses the dataset from movielens, link: http://grouplens.org/datasets/movielens/100k/

The approach used is MatrixFactorization using Stochastic Gradient Descent. First, the given movielens dataset is used to create a movie-movie similarity matrix (using Collaborative Filtering). Then this data is used to display the top similar movies to a movie when the user browses to the particular movie.

It also trains the users on the go, i.e. when the users rate movies, these data are used to refine the recommendations. For this, the users are again trained separately.

Concepts Used

Cosine Similarity
Matrix Factorization
Stochastic Gradient Descent
MovieDB API Usage for Movie Images

Reference Links:

http://blog.ethanrosenthal.com/2016/01/09/explicit-matrix-factorization-sgd-als/
https://www.youtube.com/watch?v=UfNU3Vhv5CA

Usage:

Get the necessary packages:

numpy, pandas, matplotlib

Migrate the database:

./manage.py makemigrations
./manage.py migrate

Populate the database with movie details (takes a few hours, if you have a better way, please suggest)

cd recommendations/
from populate import *
AllMovies.get_all_movies()
AllCritics.get_all_critics()
AllUsers.get_all_users()
AllRatings.get_all_ratings() 
AllGenres.get_all_genres()