Homepage Inria website

Section: New Results

Querying Heterogeneous Linked Data


The computation of the provenance of a query answer is a classical problem in database theory. It consists in aggregating the impact of tuples of a database to a query answer. This allows to give an explanation of the query answers, that can help to judge their reliability. The computation of the provenance of a query answer is thus an aggregation problem as studied by the ANR project Aggreg .

P. Bourhis [20] showed at PODS — the top conference on database theory — that the lineage of MSO queries on treelike database instances is tractable, but not on other instances. This work was in cooperation with Telecom ParisTech and ENS Paris. As a first application, he can show that MSO query evaluation on probabilistic databases is tractable for tree like database instances, but not otherwise.

P. Bourhis applied in cooperation with Tel Aviv, provenance problems to recommendation systems. This allows to explain the end result by summarising with similar data without changing significantly results obtained in general by aggregation on the data. The corresponding tool was demonstrated at EDBT [32].

Certain Query Answering and Access Control

The problem of certain query answering consists in finding which are the certain answer of a query in a database with incomplete data, and a set of constraints representing available the knowledge on the incomplete data.

P. Bourhis [24] presented at LICS — the top conference in logic in computer science — a general framework for querying databases with visible and invisible relations. This work was done in cooperation with Oxford, Santa Cruez, and Bordeaux. His framework is motivated by the problem of access control for relational databases, i.e. of data leakage in relational views, but generalizes at the same time the problem of certain query answering. Invisible relations are subject to the open world assumption possibly under constraints as usual in certain query answering, while visible relations are subject to the closed world assumption. Bourhis then show that it is decidable, whether a conjunctive has an answer in this framework, when given the visible relation, the constraints, and the query as inputs. He also studies the complexity of this problem. It turns out the complexity increases from polynomial to doubly exponential, compared to certain query answering, since adding visible relations subject to the closed world assumption.

P. Bourhis studied at IJCAI [19] certain query answering with some transitive closure constraints, which allow to define a constraints with recursion. This work was done in collaboration with Oxford and Telecom ParisTech.

The problem of ontological query containment consists in establishing whether the certain answers of two queries subject to an ontology are included in each other. P. Bourhis [26] studied at KR this problem for several closely related formalisms: monadic disjunctive Datalog (MDDLog), MMSNP (a logical generalization of constraint satisfaction problems) and ontology-mediated queries (OMQs). This work was done in cooperation with Bremen.

Recursive Queries

At LICS [21] again, P. Bourhis showed in collaboration with Oxford how to lift a major restriction on decidable fixpoint logics that can define recursive queries (such as C2RPQs), specifically on guarded logic. This allows to improve significantly expressiveness of decidable fixpoint logics.

A. Lemay contributed at TKDE [14] the gMark benchmark, a tool to generate large size graph database and an associated set of queries. This work was done in cooperation with Eindhoven and previous members of Links that are now in Lyon and Clérmont-Ferrant. The tool was also demonstrated at VLDB [13]. Its main interest is a great flexibility (the generation of the graph can be done from a simple schema, but can also incorporate elaborate a parameters), an ability to generate recursive queries, and the possibility to generate large sets of queries of a desired selectivity. This benchmark allowed for instance to highlight difficulties for the existing query engines to deal with recursive queries of high selectivity.

Data Integration

P. Bourhis and S. Staworko in cooperation with Bordeaux and Oxford presented at TODS [17] their work on bounded repairability for regular tree languages, which is a study on whether a tree document (typically XML) can be repaired to fit a given target tree language within a bounded amount of tree editing operations. The article studies the complexity of different classes of tree languages such as non-recursive DTDs, recursive DTDs, or languages by arbitrary bottom-up tree automaton.

J.M. Lozano started his PhD project under the supervision of I. Boneva and S. Staworko. His topic subscribes the ANR project Datacert on data integration and certification.

Schema Validation

A. Boiret, V. Hugot and J. Niehren studied schemas for JSON documents in Information and Computation [15]. This work was done in collaboration with Paris 7. A JSON document is an unordered data trees, so schemas for such documents are best seen as automata for unordered data trees. The paper generalizes several previous formalisms for automata on unordered trees in a uniform framework. Whether the equivalence of two schemas can be tested in P-time is studied for various instances of the framework.

This work subscribes to the ANR project Colis where unranked data trees are used as models of linux file systems. In this context, N. Bacquey started his postdoc on the verification of linux installation scripts.