Fri schreiber key_knowledge engineering
- 1. Knowledge engineering and the Web Guus Schreiber VU University Amsterdam Computer Science, Network Institute
- 2. Overview of this talk • Web data representation – a meta view • Knowledge for the Web: categories – key sources, alignment, conversion • Using knowledge: visualization and search
- 3. My journey knowledge engineering • design patterns for problem solving • methodology for knowledge systems • models of domain knowledge • ontology engineering
- 4. My journey access to digital heritage
- 5. My journey Web standards Chair of • RDF 1.1 • OWL Web Ontology Language 1.0 • SKOS model for publishing vocabularies on the Web • Deployment & best practices
- 6. A few words about Web standardization • Key success factor! • Consensus process actually works – Some of the time at least • Public review – Taking every comment seriously • The danger of over-designing – Principle of minimal commitment
- 7. Example: W3C RDF 1.1 group • 8K group messages (publicly visible) • 2K messages about external comments • 125+ teleconferences • 200 issues resolved
- 9. Caution • Representation languages are there for you • And not the other way around ….
- 10. HTML5: a leap forward Rationale • Consistent separation of content and presentation • Semantics of the structure of information Typical new elements <article>! <section>! <aside>! <header> <footer>!
- 11. RDF: triples and graphs RDF is simply labeling resources and links
- 13. RDF syntaxes • Human-readable: Turtle/TriG • Line-based: N-Triples/N-Quads • HTML embedding: RDFa • JSON-based: JSON-LD • XML-based: RDF XML
- 14. Data modeling on the Web RDF • Class hierarchy • Property hierarchy • Domain and range restrictions • Data types • OWL • Property characteristics – E.g., inverse, functional, transitive, ….. • Identify management – E.g., same as, equivalent class • …….. I prefer a pick-and-choose approach
- 15. Writing in an ontology language does not make it an ontology! • Ontology is vehicle for sharing • Papers about your own idiosyncratic “university ontology” should be rejected at conferences • The quality of an ontology does not depend on the number of, for example, OWL constructs used
- 16. Rationale • A vocabulary represents distilled knowledge of a community • Typically product of a consensus process over longer period of time Use • 200+ vocabularies published • E.g.: Library of Congress Subject Headings • Mainly in library field SKOS: making existing vocabularies Web accessible
- 17. The strength of SKOS lies in its simplicity Baker et al: Key choices in the design of SKOS
- 18. Beware of ontological over- commitment • We have the understandable tendency to use semantic modeling constructs whenever we can • Better is to limit any Web model to the absolute minimum
- 19. KNOWLEDGE ON THE WEB: CATEGORIES
- 20. The concept triad Source: http://www.jamesodell.com/Ontology_White_Paper_2011-07-15.pdf.
- 21. Categorization • OWL (Description logic) takes an extensional view of classes – A set is completely defined by its members • This puts the emphasis on specifying class boundaries • Work of Rosch et al. takes a different view 21
- 22. Categories (Rosch) • Help us to organize the world • Tools for perception • Basic-level categories – Are the prime categories used by people – Have the highest number of common and distinctive attributes – What those basic-level categories are may depend on context 22
- 24. 24 FOAF: Friend of a Friend
- 25. 25 Dublin Core: metadata of Web resources
- 27. schema.org categories for TV programs
- 28. schema.org issues • Top-down versus bottom-up • Ownership and control • Who can update/extend? • Does use for general search bias the vocabulary?
- 29. The myth of a unified vocabulary • In large virtual collections there are always multiple vocabularies – In multiple languages • Every vocabulary has its own perspective – You can’t just merge them • But you can use vocabularies jointly by defining a limited set of links – “Vocabulary alignment”
- 30. Category alignment vs. identity disambiguation • Alignment concerns finding links between (similar) categories, which typically have no identity in the real world • Identity disambiguation is finding out whether two or more IDs point to the same object in the real world (e.g., person, building, ship) • The distinction is more subtle that “class versus instance”
- 31. Alignment techniques • Syntax: comparison of characters of the terms – Measures of syntactic distance – Language processing • E.g. Tokenization, single/plural, • Relate to lexical resource – Relate terms to place in WordNet hierarchy • Taxonomy comparison – Look for common parents/children in taxonomy • Instance based mapping – Two classes are similar if their instances are similar.
- 34. Be modest! Don’t recreate, but enrich and align • Knowledge engineers should refrain from developing their own idiosyncratic ontologies • Instead, they should make the available rich vocabularies, thesauri and databases available in an interoperable (web) format • Techniques: learning, alignment
- 35. 35
- 36. Issues in conversion to Linked Data • Keep original info as much as possible • Ensure standard datatypes for value conversion • Choice of concept URL • Vocabulary URL: our strategy – Large organizaCons: use their own – Small organizaCons: purl.org URL 36
- 38. Winner EuroITV Competition Best Archives on the Web Award http://waisda.nl 38
- 39. http://waisda.nl http://www.prestoprime.org/ @waisda Improve video search for (1) fragment retrieval & (2) within video navigation by using crowdsourcing for (1) including time-based annotations & (2) bridging the vocabulary gap of searcher & cataloguer
- 40. Nichesourcing: finding the right minority vote “the frog stands for the rejected lover” “The frog indicates the profession of the women” “Sexy ladies!” annota;on quality level 40 “women”
- 42. USING KNOWLEDGE: VISUALIZATION AND SEARCH
- 45. Extracting piracy events from piracy reports & Web sources
- 46. Enriching description of search results
- 47. Sample graph search algorithm From search term (literal) to art work • Find resources with matching label • Find path from resource to art work – Cost of each step (step when above cost threshold) – Special treatment of semantics: sameAs, inverseOf, transitive, …. • Cluster results based on path similarities
- 48. Graph search
- 49. Example of path clustering Issues: • number of clusters • path length
- 50. Using alignment in search “Tokugawa” SVCN period Edo SVCN is local in-house ethnology thesaurus AAT style/period Edo (Japanese period) Tokugawa AAT is Getty’s Art & Architecture Thesaurus
- 51. Loca;on-‐based search: Moulin de la GaleCe relatively easy
- 52. Relation search: Picasso, Matisse & Braque
- 54. Acknowledgements • Long list of people • Projects: COMMIT, Agora, PrestoPrime, EuropeanaConnect, Poseidon, BiographyNet, Multimedian E-Culture