[Objective] The relationship of the knowledge graph in the financial field changes frequently, and the traditional relationship prediction method relying on manual annotation cannot meet the needs of the financial field. Therefore, a data driven dynamic relationship prediction method is proposed to provide a new perspective for the research on the rapid updating method of financial knowledge graph. [Methods] The model crawls relevant information on the Internet on time according to the monitoring list, uses the Mask Language Model to build the dataset and train the model. The hierarchical structure of the financial knowledge graph is extracted to build a hidden layer of the neural network. The neurons contained in the hidden layer represent named entities. The hidden layers are connected by a relationship matrix. The dynamic prediction of relationships is achieved by updating the connection matrix. [Results] Taking the two equity changes at the beginning of the "Baowan" event as an example, the proposed method can quickly capture the changes in the relationship between corresponding entities in the financial graph in different periods, which verifies the effectiveness of the proposed method. [Limitations] Due to the characteristics of unsupervised learning, the predicted relationship is relatively divergent, and manual calibration verification is still required. [Conclusions] The experimental results show that the proposed method can effectively capture the changes in the relationship between entities without manual annotation under the condition of sufficient data. The relationship between financial knowledge graph can be effectively and continuously predicted.