In order to more accurately capture the evolutionary patterns of disaster risks and facilitate scientific risk assessment, this study innovatively integrated the dual natural and social attributes of disasters. By leveraging the outcomes from event extraction, event relationship identification, and event generalization, a method based on causal association strength was developed to mine event logic. Subsequently, an emergency event evolutionary graph was constructed for assessing chain-reaction disaster risks. Combined with a similarity matching method that incorporates the multi-dimensional attributes of disaster events, an empirical study was conducted using earthquake disasters as a case study. The results demonstrate that the proposed disaster risk assessment framework based on the event evolutionary graph can rapidly identify historical earthquake events with high similarity to the Luding earthquake case. Furthermore, by leveraging the intrinsic interconnections and evolutionary paths among risks revealed by the graph, the framework can predict secondary disasters (e.g., landslide-dammed lakes) and derivative social risks (e.g., rumor propagation).