Integrating knowledge graph, complex network and Bayesian network for data-driven risk assessment

Bayesian network is an effective method for quantitative risk assessment, but most existing studies are either heavily data-dependent or excessively expert-dependent. In this paper, knowledge graph, complex network theory and Bayesian network are integrated into a KCB model for data-driven risk assessment, especially small data situations. By applying knowledge graph with natural language processing, a causation graph could be extracted and illustrated from accident reports. Some indexes from complex network theory are introduced to identify critical nodes to simplify the huge graph. Based on the simplified network, a Bayesian network is established to quantitatively demonstrate accidents from causes to consequences. Moreover, sensitivity analysis and scenario analysis are conducted to support the decision-making of safety management. In all, the expert involvement of Bayesian network can be reduced by applying the KCB model. Besides, the KCB model can be further applied to many other areas to reach uncertainty modelling.