Big data ontology_summit_feb2012
Download as PPTX, PDF1 like427 views
1) Ontology technology can help integrate big data by annotating labels from different databases with terms from common controlled vocabularies, providing benefits for search, integration, and potential reasoning. 2) However, this approach often fails because it is easy to create many incompatible ontologies in siloed ways, undermining semantic integration goals. 3) For ontologies to succeed, we need an incremental process where good ontologies survive and spread, and people are incentivized to reuse high-quality, tested ontologies rather than creating new ones.
Big data ontology_summit_feb2012
- 1. Big Data that might benefit from ontology technology, but why this usually fails Barry Smith National Center for Ontological Research 1
- 2. The strategy of annotation Databases describe data using multiple heterogeneous labels If we can annotate (tag) these labels using terms from common controlled vocabularies, then a virtual arms- length integration can be achieved, providing • immediate benefits for search and retrieval • a starting point for the creation of net-centric reference data • potential longer term benefits for reasoning with no need to modify existing systems, code or data See Ceusters et al. Proceedings of DILS 2004. http://ontology.buffalo.edu/bio/LinkSuite.pdf 2
- 3. String searches yield partial results, rest on manual effort and on familiarity with existing database contents Ontologies facilitate grouping of annotations brain 20 hindbrain 15 rhombomere 10 Query ‘brain’ without ontology 20 Query ‘brain’ with ontology 45
- 4. Examples of where this method works • Reference Genome Annotation Project http://www.geneontology.org/GO.refgenome.shtml • Human resources data in large organizations http://www.youtube.com/watch?v=OzW3Gc_yA9A • Military intelligence data Salmen et al. in http://ceur-ws.org/Vol-808/ Other potential areas of application: • Crime • Public health • Insurance • Finance 4
- 5. But normally the method does not work Semantic technology (OWL, …) seeks to break down data silos Unfortunately it is now so easy to create ontologies that myriad incompatible ontologies are being created in ad hoc ways leading to the creation of new, semantic silos The Semantic Web framework as currently conceived and governed by the W3C (modeled on html) yields minimal standardization The more semantic technology is successful, they more we fail to achieve our goals 5
- 6. Reasons for this effect • Just as it’s easier to build a new database, so it’s easier to build a new ontology for each new project • You will not get paid for reusing existing ontologies (Let a million ontologies bloom) • There are no ‘good’ ontologies, anyway (just arbitrary choices of terms and relations …) • Information technology (hardware) changes constantly, not worth the effort of getting things right 6
- 7. How to do it right? • how create an incremental, evolutionary process, where what is good survives, and what is bad fails • create a scenario in which people will find it profitable to reuse ontologies, terminologies and coding systems which have been tried and tested • silo effects will be avoided and results of investment in Semantic Technology will cumulate effectively 7
- 8. Uses of ‘ontology’ in PubMed abstracts 8
- 9. By far the most successful: GO (Gene Ontology) 9
- 10. GO provides a controlled vocabulary of terms for use in annotating (tagging) biological data • multi-species, multi-disciplinary, open source • built and maintained by domain experts • contributing to the cumulativity of scientific results obtained by distinct research communities • natural language and logical definitions for all terms to support consistent human application and computational exploitation • rigorous governance process • feedback loop connects users to editors 10
- 11. How to do it right • ontologies should mimic the methodology used by the GO (following the principles of the OBO Foundry: http://obofoundry.org) • ontologies in the same field should be developed in coordinated fashion to ensure that there is exactly one ontology for each subdomain • ontologies should be developed incrementally in a way that builds on successful user testing at every stage 11