With the development of LBSN, more and more attention has been paid to constructing interest point recommendation based on historical data of interest points, and there are many algorithms to predict and analyze interest points. However, in the traditional POI recommendation system, the user's information selection is not focused, which leads to the poor utilization of user data. To solve this problem, this paper improves the existing POI recommendation algorithm DAN-SNR, and constructs a new POI recommendation algorithm model MG-DAN-SNR based on deep neural network. The simulation experiment on Foursquare data set shows that the improved model has a significant improvement in the evaluation index, and can better recommend for users.