The text records of urban rail transit operation emergencies contain information about the relationship between events, which can provide auxiliary support for emergency decision-making. However, most of the current text records of urban rail transit operation emergencies are unstructured data, which makes it difficult to extract the correlation of events directly. Therefore, this paper proposes an effective approach to extract the event relations from the text records of urban rail transit operation emergencies by using natural language processing technology. Since the research object of this paper is Chinese semantics, the characteristics of the Chinese language are considered to make a better performance. Specifically, since the shortcomings of insufficient information representation and single perspectives in the expression of word vector, this paper combines Chinese word vector with Chinese character vector to obtain the information of the text data from different levels. Then, the obtained information is input into the bidirectional long short-term memory neural network-attention mechanism model (BiLSTM-Attention), which makes the overall and local features of the text can be simultaneously extracted. Finally, the experiments illustrate the model BiLSTM-Attention proposed in this paper can effectively and accurately extract the relationship between events, which provides an effective reference for emergency decision-making.