Based on Neural Collaborative Filtering paper for recommending movie and Pinterest posts according to user interactions.
The model is built on PyTorch and combines the more simple Matrix Factorization and the more complex Multi-Layer perceptron to learn both linear and non-linear relationships between user interactions. It was made clear by the paper that this results in better HR@10 (Hit Rate @ 10) and NDCG@10 (Normalized Discounted Cumulative Gain @ 10) scores.
The model was able to achieve a HR@10 of almost 0.7 and a NDCG@10 of around 0.4 which aligns with the scores found in the research paper.