Hobby project for learning Apache Spark. I implemented a recommendation engine using Apache Mahout a while ago. Mahout uses Item-Item similarities in it's collaborative filtering algorithm. I'd like to test the same in Spark as it uses a different collaborative filtering algorithm.
Example data can be downloaded from: https://grouplens.org/datasets/movielens/
/train trains the model using a csv dataset with format userid, productId, rating
Training should take less than a minute with the small dataset at: https://grouplens.org/datasets/movielens/
/recommend/userId returns personilized recommendations for the user
Just started this project and there's a lot of stuff to do:
- Verification of training result
- Persisting model (at the moment recommendation are not instantaneous)
- Item based recommendations
- Custom number of recommendations