基于时序知识图谱推理的网络极端化风险行为预警研究OACHSSCD

[Purpose]To enhance the capability of identifying and providing early warning of online extremism risk behaviors in public se-curity intelligence analysis,this study aims to uncover the temporal patterns underlying individuals'evolving tendencies toward extremism expressed in cyberspace.[Method]This study develops an early-warning framework for online extremism risk behaviors based on tempo-ral knowledge graph reasoning and proposes HOO-TKE,a temporal knowledge graph extrapolation model featuring an offline-online memory-fusion reasoning mechanism.The model consists of an offline knowledge reasoning module,an online dynamic reasoning mod-ule,and a historical memory integration module.By leveraging cross-temporal feature fusion and multi-level semantic modeling,the pro-posed approach addresses the limitations of traditional models in capturing long-and short-term dependencies and ensuring semantic conti-nuity in behavioral evolution.Experiments are conducted on three internationally recognized temporal event datasets widely used for fore-casting social conflicts and risk events.[Result/Conclusion]Experimental results show that HOO-TKE achieves an average improvement of 1.96%in MRR,2.17%in Hits@1,2.07%in Hits@3,and 1.08%in Hits@10,demonstrating the effectiveness and interpretability of the model in temporal knowledge reasoning and risk behavior early warning.