Knowledge warehousing with WebKB-2

A knowledge server for structuring, sharing and publishing information


Dr Philippe MARTIN

e-mail: pm .@. phmartin dot info




Printable slides accessible at http://www.webkb.org/doc/slides/dstc/













Plan














Documents, databases, knowledge bases


  • The semantic content of documents cannot be extracted automatically

  • Lexical techniques (e.g. keyword-based techniques) do not support question answering (-> precision/recall ratio, redundancies, no structure, statistics instead of inferences)

  • Data indexation and retrieval in databases and structured documents must follow predefined schemas

  • Knowledge bases (KBs) permit to store and interconnect logical representations of facts, rules or definitions according to dynamically updatable hierarchy of concepts and relations (ontologies), and thus may support knowledge sharing, comparison and retrieval via conceptual browsing or querying

But knowledge representation is a difficult manual task that is not much facilitated by current knowledge base systems/servers.













Data/knowledge warehouses


Merging data from distributed databases require specially-designed wrappers (+ middlewares such as CORBA) and may be inefficient. In such a case, data warehousing is the solution.

Correctly merging knowledge representations from independently developped KBs is much more difficult, even manually.
Genuine knowledge sharing implies knowledge warehousing.













The advantages of WebKB-2


  • Can be used by Web users/agents (www.webkb.org)

  • Supports very large KBs

  • Initialized with the biggest general KB after CYC

  • Exploits the KB to generate forms for guiding and easing knowledge entering

  • Has intuitive expressive and high-level input/output formats

  • Permits users to update a shared KB without lexical/semantic conflicts nor (detected) redundancies, and without obliging the users to agree with each other.

Applications:

  • Yellow-Pages, Auctions, Classifieds, Shopping, Jobs, Personals, ...

  • Corporate memories, cooperative ontology building, state of the art












Examples with generated interfaces


  • Looking for the price of used Toyota Corolla manual cars in Australia

  • Looking for used cars for sell in Southport for at most 7000 AUD

  • Looking for new cars for sell by a car_dealer













Examples with a tailored interface


  • Locating short-term accomodations and attractions on the Sunshine Coast

  • Locating hotels or bed-and-breakfast charging less than 100 AUD a night, where check-out is no sooner than 11:00, without a minimum stay of 3 nights, near a Chineese restaurant

  • Locating Eurasian restaurants near the selected accomodation

  • Locating Chineese restaurants serving seafood












More examples of knowledge representation/sharing














Summary




These slides are accessible at http://www.webkb.org/doc/slides/dstc0302.html