Temporal Relational Knowledge Graph Construction for Hot Event News

Hot events are public accident that occur suddenly at a certain time, trigger extensive news media coverage, and may have significant impacts on public safety and social stability. Exploring the temporal relations and intrinsic argument associations of bursty hot events is crucial for understanding the mechanisms of occurrence and development of event, as well as for event analysis and prediction task. Knowledge graph is a powerful tool of knowledge representation that characterizes entity attributes and their relations. However, it is challenging for existing method to model the relation of temporal development and the dynamic associations of entities of hot event news at multiple granularities. We propose a knowledge graph construction framework for hot event news, which jointly constructs multi-dimensional event knowledge representations by integrating event cluster statistical relation, entity dynamic relation, and event causal relation. We propose an event entity dynamic relation extraction model that incorporates the syntactic determination of composite arguments. Finally, the effectiveness and practicality of the proposed method are validated on real news data.