This application creates a web dashboard that could be used to generate movie recommendations for groups. It also allows you to tune your own recommendation model selecting it's hyperparameters.
Video in Spanish showing the application
This application uses python 3, pip and you need to install the dependencies inside file requiremetns.txt
- Install all application requirements using pip(you need to have pip installed in your computer)
pip install -r requirements.txt
- Get into the src folder and run the app using python
python3 -m src.app
After starting the app, a web dashboard should be available at http://127.0.0.1:8050/. That web page shows basic data insights on the root page and if you go to the url http://127.0.0.1:8050/recommendations, you can train a recommendation model selecting his hyper-parameters.
Once the model is trained, you can select a group of users from the the table that is located under the model hyper-parameters. Another table will show selected users performed ratings. Once you are done creating a group, you can click on the button GET RECOMMENDATIONS FOR USERS to generate group recommendations, which will be loaded into another table at the end of the page
This aplication uses movielens dataset and it is located under the folder data, that contains the following files:
movies.csv: contains 193609 different movies with the feilds movie id, title and genre for all available moviesratings.csv: contains 100836 ratings performed by users and each rating contains the user id, timestamp and rating performed by each user to a movie
Matrix factorization is performed using a self made implementation of the SVD++ algorithm, wich is used to perform matrix factorization on the ratings matrix.
Implementation can be found on the train method of the file src/services/SGD.py
Group recommendation are obtained using the method/criteria Least Misery Criteria over the individual recommendation of each user inside a group. Impementation is available on the method obtain_group_recommendations inside the file src/services/SGD.py
More details about the recommendation process and used data can be found inside this google colaboratory notebook. You can read an offline rendered version of that notebook on one of the following files: