Answers are grounded in this paper and cite their sources.
Railway operational hazards are interrelated and can cause accidents through cascading effects. Identifying precursor hazards and predicting subsequent hazards are necessary to prevent railway operational accidents. Knowledge graphs (KGs) explicitly encode entity relations and suit modelling such interrelated hazards. However, existing knowledge graph embedding (KGE) methods are not designed to handle combined link types induced by hazard interactivity. To address this, a novel KGE-based approach is proposed to infer both causes and effects of a target hazard from historical data. It is evaluated on a real-world railway operational hazard knowledge graph (ROHKG) built from 353 UK Rail Accident Investigation Branch (RAIB) accident reports over 2005–2023, with 948 hazard events and 1897 causal relation triples. Two novel KGE models are developed, namely a multiple semantic translation model for cause identification and a bidirectional semantic translation model for effect prediction. Both jointly integrate semantic correlation and translational distance features to mitigate mutual interference of combined link types. The models achieve Hit@10 of 84.45% and 90.24%, and mean reciprocal rank (MRR) of 0.4464 and 0.5102, for cause identification and effect prediction, respectively. They outperform leading baselines by 26.03% in MRR and 27.32% in Hit@10 for cause identification, and by 20.02% in Hit@1 for effect prediction. Both remain applicable when training data is reduced to 60% and 70% of the original corpus. These results confirm the approach’s effectiveness and robustness under data-limited conditions, providing railway operators with a basis for prioritising hazards for further assessment.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: