Section: New Results

Business Intelligence

Participants : Corentin Follenfant, Olivier Corby, Fabien Gandon.

This PhD Thesis is done with a CIFRE industrial grant from SAP Research.

Industrial Business Intelligence proposes tools and methods to perform data analysis over heterogeneous enterprise sources. They allow one to harvest, federate, cleanse, annotate, query, organize and visualize data in order to support decision making with human-readable documents such as reports, dashboards, mobile visualizations. Authoring these dynamic documents requires proficiency in technical domains like relational modeling and SQL for one to produce relevant content: end users therefore praise example-driven and information retrieval (IR) systems that help them reusing existing content. Such systems need common structured metadata to enable comparison, search, matching and recommendation of (parts of) documents.

As target data sources are mainly tabular or relational, queries executed to feed the dynamic documents are SQL or derivatives. In [62] we proposed to model these queries as RDF named graphs, and use the graphs as documents annotations. Queries are represented through their abstract syntax trees (AST) represented with RDF graphs. The SQL-specific modeling contribution can therefore be applied to any generic query language. We identified two desirable features for IR systems that deal with queries repositories: search and rewriting, the latter allowing further annotation as well as reconciliation of source relational entities against LOD (Linked Open Data) repositories. On this basis we evaluated SPARQL 1.1 to perform SQL query analysis, i.e. pattern-matching search or rewriting, using in particular property paths. Resulting SPARQL queries are intuitive and concise.

Next steps include a quantitative evaluation by extracting RDF representations from a repository of SQL-fed documents, the production of a library of SPARQL queries that perform generic IR operations against RDF-modelled SQL queries, a formalization of the modeling and operations to compare them with generic tree manipulation methods. In further work we plan to investigate rewriting queries from different languages modelled with language-specific abstract syntax trees to generic abstract syntax trees and experiment cross-language query comparison with SPARQL.