- From: Paul Tyson <phtyson@sbcglobal.net>
- Date: Sun, 12 Oct 2014 21:02:52 -0500
- To: Paul Trevithick <ptrevithick@gmail.com>
- Cc: Semantic Web <semantic-web@w3.org>
On Thu, 2014-10-09 at 17:28 -0400, Paul Trevithick wrote: > I'm working on project that involves integrating data from many > disparate sources into an RDF repository-based data warehouse. We plan > to develop an OWL ontology for data in the repository. We plan to code > up data connectors to perform ETL operations and populate the > repository with RDF data that adheres to this common ontology. The > data sources use various access methods, data formats and schemas. I discovered that the first step ("develop an OWL ontology") is unnecessary for practical business applications. Use the least-effort RDB2RDF mapping, which means something like the "direct mapping" of the RDB2RDF spec. The folks who developed the "disparate" database schemas have already covered all the ground you could possibly cover with an ontology design effort. "Integrate" the data nearer the point of use by SPARQL queries; or, to capture significant business logic, write RIF rules that can be translated to SPARQL. If it helps, enrich the repository with SKOS assertions or other materialized predicates for more complicated business logic that can't be converted to interactive run-time queries. This approach means you will spend much less time on the "universal" ontology and much more time actually making disparate data sources usefully accessible. That means writing queries and rules. This effort is greatly aided by having ready documentation of all schemas available, which pretty much falls out of the box if you keep stuff in RDF and XML. Best of luck, --Paul > > In the non-semweb world there are technologies that allow the "source" > and the "target" schemas to be defined, that allow declarative rules > to be developed (sometimes by point and click tools) the source schema > to the target schemas, and code that consumes these rules and > transforms the data into the target schema. In the non-webweb world > our data connectors could leverage these technologies. > > Are there emerging best practices for doing the equivalent of schema > mapping in semweb world? We'd like to limit our data connector code to > generating RDF triples from a data source, and rely on some other > technology to, given some kind of ontology-to-ontology mapping rules, > transform RDF data adhering to the source ontology into RDF data > adhering to the repository's target ontology. > > Paul
Received on Monday, 13 October 2014 02:05:48 UTC