Quarser: a graph-aware JSON-LD parser

Abstract

The continuous growth of the Web of Data has fueled the interest of performing analytical operations over Knowledge Graphs (KGs). The challenge of handling large scale KGs foster the research on optimization and benchmarking of existing Semantic Web solutions. Most of them focus on query planning in the context of one-time queries. Nonetheless, the spreading of application domains like Internet of Things (IoT), Social Media Analytics, and News Analysis has focused attention on different kinds of queries that tend to be recurrent. Our focus is on the performance optimization of recurrent analytical SPARQL queries by leveraging the computation spent on the parsing process of data. The literature on this type of optimization in SQL workloads is being recently explored with positive results. To the best of our knowledge, in the Semantic Web landscape, the effort has been minimal. The current thesis presents a new JSON-LD parser called Quarser, that is particularly tailored to this class of applications. Quarser is aware of the RDF graph that the parser traverses and shares the same space to compute SPARQL variable bindings. Our results, tested over the LUBM Benchmark, show a reduction of 20% of the total time of query-answering.