A machine learning-based movie recommendation system built with TensorFlow. It uses a deep neural network to learn the user's movie preferences and predicts ratings for movies the user has not yet seen.
- Connect to a TMDb account and use the list of rated movies as labelled input data
- Connect to IMDb and use a CSV-formatted file containing labelled movies as input data
- Interactive command-line interface
- Predict movie ratings (between 0 and 9) for any movie that can be found on IMDb (or TMDb, depending on the backend)
Due to some of the dependencies, this project requires Python 2.7. I installed the following into a dedicated virtualenv:
- TensorFlow for Python 2.7
- tmdbsimple
- imdbpy
- pandas
- progressbar2
- scikit-learn
If applicable, activate the virtualenv, cd into the movie_recommender directory inside the repository and
launch python movie_recommender.py. The CLI will guide you through everything else.
- Feature inference using an autoencoder. This can considerably reduce the dimensionality of the feature space and improve prediction accuracy.
- Test and improve model serialization/deserialization.
- Functionality for updating the model incrementally: Do not retrain the classifier from scratch if a new movie was added to the data set.