A Semantic-Gated Structural Diversity Penalty to Mitigate Orchestrated Recommendation Attacks in the SIoT

Abstract

The Social Internet of Things (SIoT) trust algorithms typically assume peer recommendations are independent, this leaves them vulnerable to coordinated Fake Relationship Attacks (FRA). Adversaries exploit this vulnerability by creating dense malicious cliques to manipulate trust calculations and bypass standard anomaly thresholds. In an effort to eliminate this issue without disrupting legitimate consensus, this paper proposes a Semantic-Gated Structural Diversity Penalty based on localized neighborhood partitioning. The defense operates purely on the edge and is able to extract connected components from the recommendation subgraph, separating the coordinated topological islands. The framework acts as a multi-stage logic gate to neutralize Sybil cliques executing trust inflation or synchronized bad-mouthing, while protecting the legitimate collaborative communities. An exponential discounting factor is applied exclusively to synchronized malicious cliques, preserving the trust weights of honest nodes. Experimental results across escalating attack volumes, adaptive jitter, and node mobility scenarios demonstrate that our mechanism systematically suppresses trust manipulation. The framework sustains accurate node classification with high resilience (demonstrating an empirical F1-Score of 0.986 and a False Positive Rate of 0.006), achieving security convergence at a lightweight O(k2) edge complexity. This approach provides robust protection against orchestrated collusion without requiring heavy cryptographic validation or intensive machine-learning-based trust estimation.