Two-Phase Similarity Feature Construction for Enhancing Sensor Knowledge Graph Alignment via Genetic Programmings

The rapid evolution of the Internet of Everything (IoE) has increased data complexity in urban traffic networks, necessitating the use of the semantic sensor Web (SSW) to integrate semantic metadata with sensor data via sensor knowledge graphs (SKGs). However, the heterogeneity of SKGs, with varying focus, terminology and structure, poses challenges for accurate sensor data analysis. To identify semantically identical entities across different SKGs, similarity features (SFs) capture entity similarity from multiple perspectives, but the multidimensional heterogeneity of SKGs prevents any single SF from being universally effective. To improve SKG alignment, this article presents a novel two-phase SKG alignment method, which consists of three new components. First, an automated SF construction framework is developed, which uses multiobjective genetic programming (MOGP) and single-objective genetic programming (SOGP) to automatically construct and combine the high-quality SFs. Second, new fitness functions are designed to guide the search direction of MOGP and SOGP, without relying on standard alignments. Lastly, lexicase crossover and mutation are proposed to adaptively enhance population diversity, ensuring high-quality SKG alignment. Experiment utilizes two KG datasets from the ontology alignment evaluation initiative (OAEI), along with ten pairs of practical IoE SKGs, were utilized to evaluate the performance of our approach. The results show that our method outperforms state-of-the-art matching methods, particularly in handling complex entity heterogeneity.