Personal project: music recommendation using topic modeling.
Since music is my only passion out of school, I like spending lots of time on reading music critics and listening to various albums. I also used those popular web music player, e.g. Spotify and iHeartRadio. However they recommend very limited songs and artists to me. For instance, if you're listening to the "Slowdive Radio" (Slowdive is a shoegaze band), you'll only be able to listen to Slowdive, Ride, MBV and those shoegaze bands. You can't explore more artists from different genres. If you want some out-of-box recommendations, then you have to do your own research, e.g. read music critics.
Therefore, I'm motivated to create a content-based recommendation algorithm which captures essense from music critics. Additionally this algorithm can pick similar songs beloning to different genres.
Basically the project consists of three components:
-
A web scrpaer extracting reviews from a music critics online publication, Pitchfork.com (Scraping_Ultimate.py)
-
Retrieving contents from last.fm API (lastfm_scrape.py)
-
Applying topic modeling and TF/IDF on review contents (topic_model.py)
The algorithm is evaluated by a couple of music lovers.
In the future, I'll collect more user data and combine this algorithm with collaborative filtering to better serve users' needs.