MatchFlix
This is still a work in progess.
MatchFlix is a recommender site for Netflix. It allows users to add their friends and get recommendations based on two profiles. I made this to explore recommendation algorithms, in-particular for combining two profiles.
The recommender system uses a content-based algorithm to match user-profiles with item-profiles. The user-profile is the mean of all the users likes represented as a vector. The similarities are then calculated using cosine-theta.
Going forward, I would like to implement a hybrid system combining content-based with collaborative-filtering. For now however I am happy that content-based is the correct approach due to the low number of users and the 'cold-start problem'.
Acknowledgements
I found this tutorial very helpful for authentication.
I used The Movie Database (TMDb) API to get images.
The Netflix data is from late 2019 and can be found on Kaggle here.
TODO
- Get more up-to-date data