Paper detail

Recursive SPARQL for Graph Analytics

Work on knowledge graphs and graph-based data management often focus either on declarative graph query languages or on frameworks for graph analytics, where there has been little work in trying to combine both approaches. However, many real-world tasks conceptually involve combinations of these approaches: a graph query can be used to select the appropriate data, which is then enriched with analytics, and then possibly filtered or combined again with other data by means of a query language. In this paper we propose a declarative language that is well suited to perform graph querying and analytical tasks. We do this by proposing a minimalistic extension of SPARQL to allow for expressing analytical tasks; in particular, we propose to extend SPARQL with recursive features, and provide a formal syntax and semantics for our language. We show that this language can express key analytical tasks on graphs (in fact, it is Turing complete), offering a more declarative alternative to existing frameworks and languages. We show how procedures in our language can be implemented over an off-the-shelf SPARQL engine with a specialised client that allows parallelisation and batch-based processing when memory is limited. Results show that with such an implementation, procedures for popular analytics currently run in seconds or minutes for selective sub-graphs (our target use-case) but struggle at larger scales.