Abstract: In recent years, Attention Mechanism (AM) has been widely used in natural language processing tasks based on deep learning. Deep learning recommendation based on attention mechanism has become a new direction in the research of recommendation system. This paper discusses the structure and classification standard of attention mechanism, and analyzes the main progress and shortcomings of the existing deep learning recommendation researches based on attention mechanism from four aspects: DNN recommendation, CNN recommendation, RNN recommendation and GNN recommendation. The main difficulties in the research are illustrated. Finally, the paper points out the future direction of deep learning recommendation including multi-feature interaction attention mechanism recommendation, multi-modal attention mechanism recommendation, hybrid recommendation for multiple deep neural networks based on attention mechanism, and group recommendation based on attention mechanism.

Key words: attention mechanism, deep learning, recommendation system