A Risk Prediction Model for Network Security Using Knowledge Graph

To reduce the probability of network risk, a knowledge graph-driven risk prediction model for network security was developed by exploring the information on network detection data. Indicators of the model were determined for security protection capability, attack threat risk, network performance anomaly index, and disaster tolerance, which were included in the index system of the developed model. The primary and secondary indicators were categorized to construct the knowledge graph. The graph attention mechanism was also integrated into the model. The entity-level attention network layer learned the attention coefficients among neighboring entities in the relationship path. The relationship of level attention network layers was obtained using new entity feature vectors and the attention coefficients. Entity feature vectors were aggregated to output risk prediction results. The experimental results showed that the model accurately predicted the level of risk in network security, reducing the number of risk events to more than 38%.