This article designs and constructs a deep prediction model that combines deep neural networks and Transformer encoders. Build a prediction model using deep neural networks and Transformer encoders. Due to the powerful feature extraction ability of deep neural networks and their ability to fuse heterogeneous data from multiple sources, they are used to process basic information about users and scenic spots. The Transformer encoder is used to process user comment data, modeling long-distance dependencies between words in sentences through position encoding and self attention at the encoding end, thereby fully mining the comprehensive features of comment data and obtaining more potential user interest features. The features of the last three networks are input into the factorization machine to further learn their potential interaction relationships, thereby improving the prediction accuracy of the model and alleviating the problem of data sparsity. Finally, the effectiveness and accuracy of this recommendation model were verified through experimental comparison with traditional recommendation models.