Accident analysis of waterway dangerous goods transport: Building an evolution network with text knowledge extraction

To clarify the risk factors and evolutionary characteristics affecting the safety of waterway dangerous goods transport, this study constructed an accident evolution network using unstructured data from accident reports. We integrated Natural Language Processing (NLP) technologies and complex network model to enhance accident analysis accuracy and efficiency. Initially, using publicly available data from the China Maritime Safety Administration and the Changjiang Maritime Safety Administration, we created a knowledge corpus on waterway dangerous goods transport accidents. This involved data preprocessing, annotation, and augmentation. We then developed a method for extracting knowledge by combining Bidirectional Encoder Representations from Transformers (BERT), Bidirectional Long Short-Term Memory networks (BiLSTM), and Conditional Random Fields (CRF) to identify accident-related entities. We employed Word2vec and K-means++ to vectorize and cluster these entities, standardizing the categories to build a complex network describing accident evolution. Finally, by analyzing the topology and robustness of the network, we uncovered the logical pathways of accidents involving the transport of dangerous goods on waterways. The results demonstrate that our methods effectively visualize risk factor interactions and their impact on accident progression, aiding in the development of preventive measures for the waterway transport of dangerous goods. •Introduced an NLP-complex network method to analyze risk factors and accident evolution in dangerous goods waterway transport.•Developed a BERT-BiLSTM-CRF model for entity extraction and applied Word2vec and K-means++ for knowledge fusion.•Constructed and analyzed an accident evolution network, revealing key paths and nodes in accident progression.