Undergraduate Thesis: Recommender Systems in PyTorch

Overview

I developed a unique recommender system model in PyTorch to give video game recommendations based on an input video game. The model is a neural network based matrix factorization model.

After our model is trained, it assigns a vector value to each video game. We use nearest neighbor search to find nearby vectors which are recommendations.

Vector Visualization

We can plot our video game vectors and see what items are close by. We can note that the games are of similar genre.

Recommendation Example

In the image below we are given a list of recommendations for the game “The Last of Us Remastered”. We can see our model provides recommendations of similar genre and gameplay.

Full Thesis

Link to full thesis

https://source.sheridancollege.ca/student_work_fast_applied_computing_theses/4/