In this century of information explosion, thanks to the music streaming services provided by Spotify, Pandora and etc., we now have access to tens of millions of digital songs online. Music is becoming part of people's life -- we can almost listen to music whenever and whatever we are doing.
However, one problem posed by the rapidly increasing amount of digital music is that it becomes harder for users to screen out the desirable songs from the massive music library. Music is short -- typically 3 to 5 minutes per song, and we are too lazy to spend 1 hour exploring in the music library just to find an "Eureka" song. Thus, it is important to come up with a well-designed model that would ease users uncertainty and anxiety when they are not sure what kind of music they want listen to next, and help filter music that the users are more likely to be interested in.
In this project, we aim to improve users' experience in music streaming services by automatically and accurately recommend new songs based on users' music history. Our goal is to closely examine the Million Playlist Dataset from Spotify, design and train a classic Collaborative Filtering and a advanced Neural-Net Embeddings models, and realize playlist-to-playlist and song-to-song recommendation.