Spectrum Recommendation in Cognitive Internet of Things: A Knowledge-Graph-Based Framework

As an intelligent approach for efficient spectrum utilization in Cognitive Internet of Things (CIoT), spectrum recommendation is interesting but largely unexplored, especially in the few-shot scenario of inference on unknown ratings. For the incompetence of data-driven approaches, we consider playing a guiding role of domain knowledge by introducing it into this few-shot inference problem in the high-frequency (HF) communication, as HF communication is a case of CIoT. Firstly, a knowledge-graph-based framework for spectrum recommendation is developed to organize and represent the knowledge of HF. The knowledge graph (KG) is modeled as an undirected graph for the convenience of extracting and reasoning knowledge. Inference on channel qualities for spectrum recommendation is further converted to the node regression problem on the undirected graph. Then, inference schemes based on graph convolutional network (GCN) are proposed to solve the node regression problem, where we transfer knowledge extracted from the complete data of the source domain to the few-shot inference scenario of the target domain. Experiments on both synthetic data and real data demonstrate the effectiveness of the proposed GCN-based inference algorithm with knowledge transfer.