What-To-Watch

What-To-Watch is a custom movie recommender system which suggests popular movies based on some custom ratings given by a user to a few movies from our huge pool. The suggestions for each user differ based on their movie choice.

The project incorporates two methods to generate recommendations. One is a purely machine-learning based method which involves collaborative filtering based on low-rank matrix factorization from scratch and uses gradient descent as an optimization algorithm. This implementation is based on Dr Andrew Ng's Youtube lectures on Machine Learning.

The second method is a purely statistical project which correlates a movie with other movies from a dataset based on ratings and popularity and shows the closest positively correlated movies for a given movie from the dataset.

The data used is provided by IMDB and is open to everyone

Results

Here is a list of movie-suggestions based on my preferences for many movies as per the first ML algorithm :

Here is a list of movie-suggestions similar to Toy Story (1995) as per the second statistical algorithm :

Installations and Setup

Use the package manager pip to install the dependencies.

pip3 install -r requirements.txt

Structure of the project

• The train.py file contains the model which needs to be trained the first time a user runs this project if the user wants the ratings to be as per his/her preferences.
• The predict.py file contains the top movies which the model recommends for the specific user
• The train.p file contains the weights used by the model after train.py has been run atleast once.
• The corr.py file contains the main script which gives us our results as per the second algorithm.
• Please explore the structure of the datasets in the movies.csv and ratings.csv files.

Usage

Algorithm 1

Please feel free to alter the ratings of the movies given in the excerpt below from train.py and experiment by adding ratings for some of your other favourite movies from the movies.csv file. While adding the movies please follow the structure I have used

On your first run after giving custom ratings to some movies in train.py run

python3 train.py

Be patient as this may take a while to train (up to 15 minutes)

After training to see the predictions run

python3 predict.py

Algorithm 2

Feel free to alter the movie for which you would like to see positive correlations. This can be done by replacing the given excerpt in the corr.py file with a movie name from the name column in the movies.csv file.

movie_user_rating = movie_matrix['Toy Story (1995)']

Make sure you copy the name exactly as given in the movies.csv file.

After choosing the movie, to see the results run

python3 corr.py

Contributing

Pull requests are welcome. For major changes, please open an issue first to discuss what you would like to change.

License

MIT