HGL-CC: Smart Home Security Risk Assessment on Device Connected-Control Graph with Hierarchical Graph Learning

In smart homes, there exist interconnected control relationships among IoT components. Complex multi-stage network attacks exploit these dependencies to gain unauthorized access or get private data. However, existing IoT security analysis methods either focus solely on the statistical characteristics of individual components or on the interactions between components. We conducted an analysis of the security architecture in smart homes and redefined the issue as a Smart Home Security Risk Assessment based on the concept of a Device Security Connected-Control Graph. Based on this framework, we introduced a Hierarchical Graph Learning model (HGL-CC) that comprises two tiers. The first tier analyzes the interconnected control relationships and vulnerability data between devices using an attack graph model. Sampling and aggregation functions are designed based on potential attack path relationships. The second tier calculates security assessment scores based on the diffusion of topological features. Adjustable thresholds on this basis also enable the generation of qualitative results. In the smart building system, we conducted qualitative and quantitative comparative experiments on the model. The model showed significant improvements in accuracy and effectiveness compared to previous methods in the evaluation. Additionally, functional tests and ablation experiments demonstrated the effectiveness of our approach.