The objective of this project was to show the efficacy of graph network models in product recommendation for small to medium-sized E-commerce providers. To do this, I introduced and benchmarked three models for recommendation: singular value decomposition (SVD), neural collaborative filtering, and graph convolutional network. Each was tested on Movielens datasets of 100k and 1M records. For benchmarking I used RMSE and my own version for ‘hit ratio’ or the ratio of users for which the model recommended items that were present in the holdout data for that user to the total number of users tested. By the hit ratio measure the graph models performed almost twice as good as the other two models on the same amount of data. In the case of the smaller dataset this came with only a minor tradeoff in training time.