[Objective] Based on the cross-modal deep learning method, the consumers' emotional expressions are analyzed through travel reviews, which identifies their sarcasm emotions.

[Methods] In this paper, we will build a cross-modal deep learning model. First, we encode multi-modal information, and extract the interaction information between text and pictures through a graph neural network. Furthermore, we adopt the attention mechanism to emphasize multi-modal features, which finally performs sarcasm detection.

[Results] Combined with travel review in Yelp conduct empirical research, and compared with related baseline models. The experimental results show that the cross-modal model proposed in this paper has certain advantages, and the accuracy of sarcasm detection achieves 88.77%.

[Limitations] The proposed model is only tested on the Hilton dataset of the Yelp website, which has not been further validated on other travel platforms.

[Conclusions] The proposed model can fully extract the interaction information between different modalities, that effectively improve the accuracy of sarcasm detection.