Benchmarking graph database backends : what works well with Wikidata?

Knowledge bases often utilize graphs as logical model. RDF-based knowledge bases (KB) are prime examples, as RDF (Resource Description Framework) uses graph as logical model. Graph databases are an emerging breed of NoSQL-type databases, offering graph operations to process and manipulate data. Although there are specialized databases, the so-called triple stores, for storing RDF data, graph databases can also be promising candidates for storing knowledge. In this paper, we benchmark different graph database implementations loaded with Wikidata, a real-life, large-scale knowledge base. Graph databases come in all shapes and sizes, offer different APIs and graph models. Hence we used a measurement system, that can abstract away the API differences. For the modeling aspect, we made measurements with different graph encodings previously suggested in the literature, in order to observe the impact of the encoding aspect on the overall performance