Semantic Matching Under Uncertainty in Event-Driven Middleware

Abstract

Publish/subscribe middleware traditionally performs syntactic filtering based on predefined attributes. However, in domains such as industrial monitoring or logistics coordination, event descriptions are frequently incomplete or context-dependent. We developed a semantic matching mechanism that translates event attributes into graph fragments aligned with a domain ontology. Rather than strict logical entailment, similarity is computed through weighted concept distance combined with rule-triggered inference. Subscription requests therefore match not only exact terms but semantically proximate ones within configurable tolerance bounds. Evaluation on a manufacturing event dataset shows that this approach increases recall without disproportionately inflating false positives. Importantly, performance remains stable under high event rates due to precomputed semantic adjacency matrices.