BGSD: A SBERT and GAT-based Service Discovery Framework for Heterogeneous Distributed IoT

Abstract

Collaboration between heterogeneous IoT (Internet of Things) devices can enrich the capabilities of the system. In scenarios where network infrastructure or cloud platforms are unavailable, enabling devices to collaborate autonomously in a distributed manner can improve the scalability and invulnerability of the system. A distributed service architecture with semantic services can enable more flexible collaboration among IoT devices, while the efficiency and accuracy of current methods on devices with poor computing and storage capacity need to be improved. Therefore, this paper proposes a SBERT (Sentence Bidirectional Encoder Representations from Transformers) and GAT (Graph Attention Networks) based Service Discovery method, called BGSD. SBERT is used to extract deep semantic features of services during offline to improve the discriminability of semantic features and reduce performance requirements for IoT devices, and GAT is used to perceive the local topology environment to improve the network navigability of service search. To evaluate the method, we generate a heterogeneous IoT topology dataset based on OWLS-TC4. The simulation results of service discovery show that BGSD has advantages in search efficiency, matching accuracy and computational cost, and can support the collaboration of IoT devices in heterogeneous distributed scenarios, which verifies the rationality of this method.