Recently, quantitative investment has become a new way of investment finance, which combines the model with the computer to predict the investment stock trend more efficiently and accurately. Stock trend forecasting is one of the difficult and important tasks in modern quantitative investment systems. With the development of artificial intelligence technology, many deep learning models have emerged. The application of deep learning in the stock market not only reduces the difficulty of stock analysis and trend prediction, but also introduces new investment methods and ideas for investors. However, most of the models are based on volume and price data, and the consideration of stock relationships are relatively simple, so we propose a Spatial-Temporal Graph Neural Network model fusing Barra factors (BSTGNN), which integrates factor knowledge, mines potential inter-stock relationships, and fully consider the heterogeneity of the relationship. The experiments conducted on the real-world stock market dataset prove that BSTGNN perform better than the baseline methods.