Neural Collaborative Filtering (NCF) has been widely used in recommendation systems, which further applies deep learning to recommendation systems. It is a general framework that can express and generalize matrix factorization and uses a multi-layer perceptron to learn the interaction function between users and items, greatly improving the performance of recommendations. However, in NCF, it treats all cross-features equally without considering the degree of influence of different features on the results. In this paper, we propose to integrate attention mechanism into NCF. Specifically, it is based on the assumption that different cross-features have different degrees of influence on the results. We consider the different impact of each feature on the model and adjust the weight of features to optimize the loss. Experimental results show that, under completely same experimental settings and datasets, the evaluation indicators have been improved compared with NCF.