Title:The Effects of Different JSON Representations on Querying Knowledge Graphs

Abstract:Knowledge Graphs (KGs) have emerged as the de-facto standard for modeling and querying datasets with a graph-like structure in the Semantic Web domain. Our focus is on the performance challenges associated with querying KGs. We developed three informationally equivalent JSON-based representations for KGs, namely, Subject-based Name/Value (JSON-SNV), Documents of Triples (JSON-DT), and Chain-based Name/Value (JSON-CNV). We analyzed the effects of these representations on query performance by storing them on two prominent document-based Data Management Systems (DMSs), namely, MongoDB and Couchbase and executing a set of benchmark queries over them. We also compared the execution times with row-store Virtuoso, column-store Virtuoso, and \mbox{Blazegraph} as three major DMSs with different architectures (aka, RDF-stores). Our results indicate that the representation type has a significant performance impact on query execution. For instance, the JSON-SNV outperforms others by nearly one order of magnitude to execute subject-subject join queries. This and the other results presented in this paper can assist in more accurate benchmarking of the emerging DMSs.