User Interaction Generation for Recommendations

Recommendation systems work by analyzing how a user interacts with items as well as the similarities between items and users. There are many ways to do this and, as with many things, neural networks are heavily utilized in contemporary techniques.

My project (GitHub and write up) showcase how one can utilize architecture taken from point cloud generation networks to statistically predict what item a user will interact with. This can then be used as a recommendation itself or can be used as augmented input to a different recommendation system, with both showing good results.