Structured incident information on transportation systems from traffic news

To extract the Knowledge Graph (KG) of incidents from unstructured traffic news, an automatic workflow is proposed in this paper integrating rule-based matching methods and the pre-trained language processing model.

Over eight thousand traffic incidents from 2016 to 2020 in Hong Kong are depicted into structured RDF (Resource Description Framework) triples by this workflow successfully and stored in the Science Data Bank. This dataset can be reused to evaluate the impact and response of historical incidents and to adjust parameters for traffic simulation. The code of this workflow can also be used to synchronously generate the KG of incidents from real-time traffic news for further proofreading and tracking of incident information.