Progressive Querying on Knowledge Graphs
Résumé
The exact evaluation of queries over knowledge graphs encoded as RDF data has been extensively studied. However, in a wide array of applications, RDF queries do not even terminate, due to performance reasons. Notably, queries on public SPARQL endpoints are oftentimes timed out without returning any results.
To address this, we propose a novel solution to the problem of progressive query answering and introduce the PING system that implements it on top of SPARK. In our approach, graph query answering leverages a hierarchical structure, which facilitates effective data partitioning, thus allowing us to reduce the sizes of intermediate results and return progressive answers. Moreover, it allows the RDF query evaluation algorithms to directly locate and access the different hierarchy levels required for query answering. Navigating through the hierarchy levels allows expanding or shrinking query results at different granularities. The extensive experimental study on real-world graph datasets, with varied query workloads, shows PING's effectiveness and efficiency, on both exact and progressive query answering, and its superiority to the most relevant baselines.
Domaines
Informatique [cs]Origine | Fichiers éditeurs autorisés sur une archive ouverte |
---|---|
Licence |