基于联邦图神经网络的网络威胁情报安全共享研究OA

[Purpose]Improving the security and privacy of cyber threat intelligence sharing is of great positive significance for strengthe-ning the construction of security intelligence systems,accelerating the in-depth release of intelligence resources,and promoting the percep-tion of threat situation intelligence.[Method]Aiming at the problems existing in the cyber threat intelligence sharing process,namely in-sufficient security assurance,limited depth of feature extraction,and mismatch between sharing schemes and structural characteristics,a Cyber Threat Intelligence Secure Sharing Model based on Federated Graph Neural Networks(CTISS-FG)was proposed.Firstly,the o-verall architecture of cyber threat intelligence sharing was modeled;secondly,Graph Neural Networks(GNN)were used to model the cy-ber threat intelligence entity relationship network and extract features embedded in complex topological structures,and the Federated Learn-ing(FL)framework and Fully Homomorphic Encryption(FHE)technology were combined to ensure the privacy and security during the sharing phase;finally,actual threat intelligence data was integrated and a confidence threshold was designed to realize intelligence sharing.[Result/Conclusion]The CTISS-FG model outperforms other baseline models in indicators such as accuracy and F1-score.On the DN-RTI Dataset,its precision,recall,and F1-score reach0.854,0.844,and0.849 respectively.This model has certain practical value for breaking information silos,promoting intelligence interaction,and unlocking highly sensitive intelligence.