(7) Metadata as Markup Overlay

(8) Surfacing Concepts

• Microformats [Microformats] are embedded into HTML Web content
  • there is no need for alternative representations
  • machines use the same resources as humans (and extract microformat data)
• Many larger Web sites are using structured back-ends
  • for pure HTML publishing, semantics are translated into HTML/CSS
  • more semantics can be represented by also producing microformats
• Microformats can also help in decentralized service scenarios
  • centralized: data has to be submitted to a hub
  • decentralized: crawlers search for microformats and use what they find
  • advantage of decentralization: loose coupling and independent control

(10) Different Representations

(11) XML as Custom Markup

• XML [../xml-fall10] can represent arbitrary (textual) data
  • works very well for tree-structured and document-style data
  • works not so well for graph-like data with no inherent order
• XML only defines a syntax for representing ordered trees
  • what the elements and attributes mean requires agreement on the vocabulary
  • XML is a good tool for data exchange, but only for the syntax part
• Agreement can be based on three different approaches
  1. everybody always uses a universally useful vocabulary (XHTML)
  2. user (groups) agree on vocabularies based on mutual interest
  3. user (groups) agree on modules and build vocabularies with these modules (UBL [http://www.oasis-open.org/committees/tc_home.php?wg_abbrev=ubl])

(13) Translation into a Universal Metamodel

(14) The Semantic Web Vision

• RDF [Microformats; Resource Description Framework (RDF) (1)] is a more general model for structured data
  • it is very fine-granular and can represent (almost) everything
  • its granularity becomes problematic in scenario with coarse granularity
• Transforming resources can be done in two basic ways
  1. using well-defined mappings between HTML and RDF (GRDDL [http://www.w3.org/TR/grddl-primer/])
  2. extracting information by analysis such as NLP [http://en.wikipedia.org/wiki/Natural_language_processing]
• Transforming resources does not require cooperation of the service
  • any service can be crawled and the resources can be transformed
• Transforming resources does not allow access to all data
  • crawling can have limits and thus the RDF data is limited as well
  • with everything as RDF the complete RDF graph could be queried

(16) Bypassing Web Publishing

(17) Trillions of Triples

• Turn the structured data from the back-end into RDF
  • the complete dataset can be transformed in one processing step
  • relationships between resources may be better preserved than by crawling
  • keeping the snapshot current can become a significant problem
• With everything in one database (a triple store [http://en.wikipedia.org/wiki/Triplestore]) queries become possible
  • SPARQL queries can query the graph of complete service data
  • large datasets easily translate into several billion triples
  • triple store implementation moves to native RDF databases
• Large triple stores can process complex SPARQL queries
  • performance is a problem with large datasets
  • RDF's simplicity bites back in the form of reification [http://en.wikipedia.org/wiki/Reification_%28computer_science%29#Reification_on_Semantic_Web]