Section: New Results

Linking Data Graphs

Learning Transformations

We consider the problem to learn queries and query-based transformations on semi-structured data from examples.

A. Boiret obtained his PhD for his work on the "Normalization and Learning of Transducers on Trees and Words" under the supervision of J. Niehren and A. Lemay. In this year, he showed how to learn top-down tree transformations with regular schema restrictions [31], [33], [34]. At LATA [22], he deepened a result of a previous PhD student of Links on learning sequential tree-to-word transducers (with output concatenation), by showing who to find normal forms for less restrictive linear tree-to-word transducers. At DLT [23], he could show in cooperation with Munich, that the equivalence problem of this class of transducers is in polynomial time, even though their normal forms may be of exponential size.

In the context of learning RDF graph transformations, S. Staworko presented a cooperation with Edinburg at VLDB [27]. Using bisimulation technique, he aims at aligning datas of two RDF Graphs that takes into account blank velues, changes in ontology and small differences in data values and in the structure of the graph. the alignement of graphs is an important first step for the inference of transformations.

Learning Join Queries

S. Staworko published in TODS an article [16] on learning join queries from user examples in collaboration with Universities of Lyon and Clermont-Ferrand that present techniques that allow the automatic construction of a join query through interaction with a user that simply labels sets of tuples to indicate whether the tuple is in the target query or not.