With the rapid development of railway transportation, railway safety problems have become more and more prominent. Previous research on railroad accident causation has mainly focused on structured data, and text data has not been fully mined and explored. This paper employed natural language processing (NLP) technology to extract 10 accident causal factors from 128 railroad accident reports. Then, their hierarchical relationships were divided through the interpretative structural modeling method (ISM), then a causal model was established based on Bayesian Networks (BN). A case study was conducted to validate the feasibility and validity of the model. The model proposed in this paper can excavate important causal factors affecting the occurrence of railroad accidents and their severity of railroad accidents. It may be of great significance for preventing accidents and improving emergency response capability, which can provide valid support for railroad safety.