Deep Learning-Based Semantic Interaction Network: Advancing IoT Data Modeling for Interoperability

This paper introduces Semantic Interaction Network (SINet), a deep learning architecture, to reshape semantic data modeling in the Internet of Things (IoT) ecosystem. SINet employs self-attention, relational reasoning, and dynamic knowledge graphs to handle IoT complexities, including sensor data, temporal dynamics, and device semantics. The dynamic graph captures semantic relationships, contextual dependencies, and evolving entity states, enabling IoT devices to engage in detailed, context-aware communication. The mathematical formulation of SINet is characterized by innovative components, including Temporal Relation Networks (TRNs) and Semantic Propagation Layers (SPLs). TRNs capture temporal dependencies within data sequences, while SPLs facilitate the propagation of semantics throughout the dynamic knowledge graph. The extensive simulations show that SINet outperforms the conventional recurrent neural network approach by approximately 6.98% and surpasses the traditional feedforward neural network by about 13.80% regarding semantic data modeling accuracy.