DL-Learner—A framework for inductive learning on the Semantic Web

Learning Commonalities in SPARQL

“…Most of the solutions exploit the (underlying) structure of the input descriptions, like trees for DL formulae (e.g., [1,14,27]) and for unary tree-shaped BGPQs [2,16], and directed single-root graphs for the RDF r-graphs of [7]. Roughly speaking, they all consist in a simultaneous traversal of the input descriptions, starting from their roots, while incrementally computing their lgg.…”

“…They rely on standard categorial graph product for the so-called simple (i.e., purely conjunctive) CGs [3], on anti-unifications of triples for general RDF graphs [9], and on anti-unifications of query heads and of query body triples for the general BGPQs considered in this paper. Further, while (some of) the above solutions take into account background knowledge in CGs, DLs, and RDF, this is not the case for the state of the art in SPARQL [2,16]: unary tree-shaped BGPQs are solely compared based on standard graph homomorphism (|= ∅ ).…”

“…Since then, it has also received consideration in the Knowledge Representation field, where least general generalizations were rebaptized least common subsumers [5], in Description Logics [1,5,14,27] and in Conceptual Graphs [3]. More recently, this problem started being investigated in RDF [7,9] and its associated SPARQL query language [2,10,16], the two prominent Semantic Web standards by W3C. Motivations.…”

“…Basic Graph Pattern Queries (BGPQs). Our definition, which we briefly outlined in [10], significantly departs from the literature by (i) considering general BGPQs, instead of unary tree-shaped ones [2,16], and crucially by (ii) taking advantage of background knowledge formalized as RDF Schema (RDFS) ontological constraints. Furthermore, to establish this definition of an lgg, we revise the standard generalization/specialization relation (a.k.a.…”

“…We provide a solution to the above problem (Section 4), which technically differs from the state of the art [2,16] in that it cannot exploit the (imposed) tree-shape of the input BGPQs to compute their lgg through a simultaneous root-to-leaves traversal. Instead, our solution traverses blindly the general (hence arbitrary-shaped) input BGPQs and builds their lgg using the notion of least general anti-unification of atoms [21,23], which is dual to the well-known notion of most general unification of atoms [22,23].…”