This project was realised by Adrien JAYAT and Ysée JACQUET.

It consists of a recommendation system for movies based on the Netflix Prize dataset published by Netflix in 2006.

Our algorithm gives a RMSE of 0.971 on the test set. In comparison, the RMSE of the Cinematch, the Netflix algorithm, was 0.9525.

Get Started

This project is using two submodules:

the Unity framework for testing.
the Doxygen tool for documentation generation, with the Doxygen Awesome theme.

First, clone the project and its submodules:

git clone https://gitlab.unistra.fr/jayat/projet-programmation-2023
cd projet-programmation-2023
git submodule init && git submodule update

Dependencies

Install make, gcc, doxygen and Zstandard compression algorithm:

Ubuntu

sudo apt update

sudo apt install make gcc doxygen zstd

Arch Linux

sudo pacman -Syu

sudo pacman -S make gcc doxygen zstd

Build

Build the project with the make command. It will generate the main executable in the current working directory.

You can run all tests with the make tests command:

make clean && make tests

How to use

After building the project, you can run the ./main executable to start the program. Use the option -h to get the following options list in your terminal.

Options

Flag	Argument	Description
`-f`	FORCE	Force to recompute all stats.
`-r`	LIKES_FILE	List of movies liked by the user.
`-n`	N	Length of the recommendation list the algorithm will give.
`-d`	DIRECTORY	The path of the folder where files corresponding to results will be saved.
`-l`	LIMIT	Forbidden to take in acount ratings with a date greater than the LIMIT.
`-s`	MOVIE_ID	Give statistics about the movie with the identifier MOVIE_ID.
`-c`	IDS	Allow to take into account only the ratings of the customers with given identifiers.
∅	NB_CUSTOMER_IDS	Number of given customer ids.
`-b`	IDS	Allow to not take into account the ratings of the customers with given identifiers.
∅	NB_BAD_REVIEWERS	Number of given bad reviewers.
`-e`	MIN	Allow to take into account only customers who rated at least MIN movies.
`-t`	TIME	Precise the executive time of the algorithme.
`-p`	PERCENT	Percentage (between 0 and 1) to quantify the importance of personnal recommendations over popular recommendations.

Note that options -r, -n, -t and -p are not used for statistics processing.

Examples

Gives the 10 best recommendations from liked movies

./main -r likes.txt -n 10

The likes.txt file contains titles of movies liked by the user. You can also give a list of movie ids directly from the command line.

./main -r 872 996 463 5582 -n 10 -p 0.8

Add the -p option to give a percentage to quantify the importance of personalized recommendations over popularity.

Gives statistics about a movie

./main -s 42

It will create a file named stats_mv_000042.csv in the stats folder, containing the min, max and average score of the movie 42.