Neural Collaborative Filtering (NCF) is widely used in recommendation systems, but it ignores the implicit information of users and project description documents and has the problem of data sparsity. In this paper, the explicit data scoring matrix is fused with the implicit feedback information of user and item description documents as input to alleviate the data sparsity problem; meanwhile, because the attention mechanism has better feature learning ability, which is beneficial to the neural network to better extract the nonlinear features of the input vector, this paper proposes a deep collaborative recommendation model based on the attention mechanism. Finally, comparative experimental analysis is conducted on Movielens and LastFM datasets, and the results show that the proposed model improves on two evaluation indexes, Hit Ratio (HR) and Normalized Discounted Cumulative Gain (NDCG), and effectively improves the recommendation accuracy.