Abstract:
Across domains such as social media, finance, healthcare, and IoT, data is increasingly produced as dynamic streams rather than as static datasets. In these settings, RDF is a natural representation model because it captures relationships explicitly, supports semantic interoperability, and enables integration across heterogeneous sources. Efficiently querying such dynamic RDF data remains challenging, however, as recomputing query results after every update is often too costly. Incremental View Maintenance (IVM) addresses this challenge by propagating changes in the data to materialized query results, thereby avoiding full recomputation. This article studies IVM for SPARQL conjunctive queries, with a particular focus on incremental join processing, which is central to SPARQL Basic Graph Pattern evaluation and especially important in graph workloads that typically require many joins. A structured analysis of the IVM approaches from related work is presented, together with a set of optimizations proposed where applicable. These approaches are evaluated in a fair and platform-independent setting by implementing them within a single incremental query engine, Incremunica. To support this evaluation, a TypeScript reimplementation of the Train Benchmark is developed and extended with shape-based queries to study the effect of query structure on performance. In addition, a dedicated Rete-specific evaluation is conducted to assess the impact of the proposed optimizations for that approach. The results show that there is no universally best incremental IVM approach, the performance depends on query shape, update type, and data characteristics. In particular, single-order delta queries perform poorly overall, higher-order delta queries are generally less competitive but outperform other approaches in some scenarios, and Rete-based approaches achieve the strongest performance in most scenarios. These findings provide practical guidance for selecting incremental join strategies for dynamic RDF workloads.
