Wander1ngW1nd/MoviesGRS_MFDP

Movies Group Recommender System

Application making personalized movie recommendations for groups of users. Based on each group member’s watch history, it provides the tailored watch list, taking into account preferences of everyone.

Demo

[image]

Installation

1\. Install git and docker engine

2\. Clone the project:

git clone https://github.com/Wander1ngW1nd/MoviesGRS_MFDP

3\. Build an image:

docker build -t movies_grs_image MoviesGRS_MFDP

4\. Run application container:

docker run --name movies_grs_app -dp 8501 movies_grs_image

5\. Figure out which port was assigned to your application:

docker port movies_grs_app

    You will see the similar output:

8501/tcp -> 0.0.0.0:<your_port_number>

6\. Go to:

http://0.0.0.0:<your_port_number>

Now you can use the app!

Usage

For illustration of the steps listed below you can watch demo

1\. Choose the size of your group (1-7 are currently supported). You will see tabs responding to each group’s member

2\. For each member of the group:

1. 1.Fill in the table inside the corresponding tab with ratings of the movies reflecting your preferences best
2. 2.The more ratings — the better
3. 3.You can add any number of movies or delete them
4. 4.You cannot add empty ratings table
5. 5.Submit your ratings
6. 6.If you want to change submitted ratings, you can just refill the table and submit it again

    You will be able to see all the submitted ratings on the bottom of the page

3\. Click Get recommendation button on the bottom of the page to get recommendations.

• —Get recommendation button will not appear until all the ratings are submitted
• —If your group size ≥ 4, it can take some time (~1 min) to make recommendations

Under-the-hood description

Application source files are structured as follows:

src
├── app.py
├── recommender.py
└── data
    ├── history_ratings.feather
    ├── movies_embeddings.feather
    └── movies.feather

app.py contains front-end part of the application made with streamlit. This is the entry-point file of the project. It defines the application interface and performs base users’ input processing. The supposed way of its usage if described in the corresponding section.

recommender.py contains the GroupRecommender class, generating recommendations based on users’ ratings. It takes the group size on instantiation, as this parameter defines the recommendation algorithm used, by the following way (described in pseudocode):

if group_size < 4:
	make_recommendation_based_on_movie_embeddings()
else:
	make_recommendations_based_on_svd_decomposition()

So, there are 2 different algorithms. They both firstly represent users and movies in the same vector space, but do it in different ways:

• —Movie-embeddings-based recommender firstly creates each group member profile. The profile is constructed as the mean of embeddings of movies watched by the user, weighted on their provided ratings. Movies embeddings were precomputed by passing their plots through pretrained paraphrase-distilroberta-base-v1 from SentenceTransformers library and saved to data/movies_embeddings.feather.
• —SVD-based recommender adds group members’ ratings to those stored in data/history_ratings.feather and performs SVD decomposition of the resulting rating matrix, receiving users’ and movies’ vector representations.

Then both algorithms take the mean of group members’ profiles as the group profile representation, and find top-10 nearest movies in the embeddings space by euclidean distance.

Certain models, parameters and usage conditions choice is described in separate repository, MoviesGRS_experiments_MFDP.

Development

Dependencies Management

Project’s dependencies are managed by poetry. So, all the dependencies and configuration parameters are listed in pyproject.toml.

To install the dependencies, follow these steps:

1\. Install git and poetry

2\. Clone the project and go to the corresponding directory:

git clone https://github.com/Wander1ngW1nd/MoviesGRS_MFDP
cd MoviesGRS_MFDP

3\. (Optional) If your python version does not match the requirements specified in pyproject.toml, install one of the matching versions

4\. Create virtual environment and activate it

poetry shell

5\. Install dependencies Though poetry provides its way for instaling dependencies, now it is recommended to use pip installation from requirements.txt:

pip install -r requirements.txt

This is motivated by the fact that scikit-surprise-1.1.3 does not support PEP-517 installation, and poetry native installation would just fail. Due to the same reason, if there are any dependencies that need to be added, it is recommended to add them through poetry add, and then update requirements.txt:

poetry add <needed_dependency>
poetry export -f requirements.txt --output requirements.txt

Pre-commit Hooks

This project uses pre-commit hooks for code quality checking. For this purpose https://pre-commit.com framework is utilized.

Now the following linters are used for pre-commit hooks:

• —black
• —isort
• —mypy
• —pylint

Their specific configuration details can be found in pyproject.toml.

To initialize pre-commit hooks, you need to execute the following command:

pre-commit install

To manually check the staged code base before committing you can run:

pre-commit run

If you want to check all files, not only those staged for commit, run:

pre-commit run --all-files

To skip some hooks while committing:

SKIP=black git commit -m "foo"