The short-term passenger flow prediction of the urban rail transit system is of great significance for traffic operation and management. The emerging deep learning-based models provide effective methods to improve prediction accuracy. However, most of the existing models mainly predict the passenger flow on general weekdays, while few studies focus on predicting the holiday passenger flow, which can provide more significant information for operators because congestions or accidents generally occur on holidays. To this end, we propose a deep learning-based model named GCN-Transformer comprising graph conventional neural network (GCN) and Transformer for short-term passenger flow prediction on holidays. The GCN is applied to extract the spatial features of passenger flows and the Transformer is applied to extract the temporal features of passenger flows. Moreover, in addition to the historical passenger flow data, social media data are also incorporated into the prediction model, which has been proven to have a potential correlation with the fluctuation of passenger flow. The GCN-Transformer is tested on two large-scale real-world datasets from Nanning, China during the New Year holiday and is compared with several conventional prediction models. Results demonstrate its better robustness and advantages among baseline methods, which provides overwhelming support for practical applications of short-term passenger flow prediction on holidays