This paper proposes a neural collaborative filtering (NCF) framework to model user-item interactions in recommendation systems using deep neural networks. The key challenge in collaborative filtering is to model the interaction between user and item features, which traditional matrix factorization (MF) approaches handle using an inner product. However, this simple interaction function may not capture the complex structure of user-item interactions. To address this, the authors introduce NCF, a general framework that replaces the inner product with a neural architecture capable of learning an arbitrary function from data. This allows NCF to generalize MF and incorporate non-linearities through a multi-layer perceptron (MLP) to model the user-item interaction function. The NCF framework is shown to significantly outperform state-of-the-art methods on two real-world datasets. The authors also demonstrate that deeper neural networks can improve recommendation performance. Additionally, they show that using a probabilistic approach with log loss and negative sampling is effective for learning from implicit feedback. The paper also presents a generalized matrix factorization (GMF) and a neural matrix factorization (NeuMF) model that combines the strengths of linear MF and non-linear MLP for modeling user-item latent structures. Experiments on the MovieLens and Pinterest datasets show that NeuMF achieves the best performance, significantly outperforming other methods. The results indicate that deep learning is effective for collaborative filtering, and that using a neural network to model user-item interactions can lead to better recommendation performance. The paper also discusses the benefits of pre-training and the importance of using appropriate negative sampling ratios for training. Overall, the study highlights the potential of deep learning in recommendation systems and provides a general framework for modeling user-item interactions.This paper proposes a neural collaborative filtering (NCF) framework to model user-item interactions in recommendation systems using deep neural networks. The key challenge in collaborative filtering is to model the interaction between user and item features, which traditional matrix factorization (MF) approaches handle using an inner product. However, this simple interaction function may not capture the complex structure of user-item interactions. To address this, the authors introduce NCF, a general framework that replaces the inner product with a neural architecture capable of learning an arbitrary function from data. This allows NCF to generalize MF and incorporate non-linearities through a multi-layer perceptron (MLP) to model the user-item interaction function. The NCF framework is shown to significantly outperform state-of-the-art methods on two real-world datasets. The authors also demonstrate that deeper neural networks can improve recommendation performance. Additionally, they show that using a probabilistic approach with log loss and negative sampling is effective for learning from implicit feedback. The paper also presents a generalized matrix factorization (GMF) and a neural matrix factorization (NeuMF) model that combines the strengths of linear MF and non-linear MLP for modeling user-item latent structures. Experiments on the MovieLens and Pinterest datasets show that NeuMF achieves the best performance, significantly outperforming other methods. The results indicate that deep learning is effective for collaborative filtering, and that using a neural network to model user-item interactions can lead to better recommendation performance. The paper also discusses the benefits of pre-training and the importance of using appropriate negative sampling ratios for training. Overall, the study highlights the potential of deep learning in recommendation systems and provides a general framework for modeling user-item interactions.