SemWalker

• A data browser, aka "semantic web browser "(cf Longwell, Tabulator)
• Demonstrates "downhill steps" to decentralizing the Data Web
• In progress

What is a Semantic Web Browser?

Software which lets users browse RDF data published on the web

• Single-Item Page
• Multi-Item Page
• Class-Specific Views
• Real-Time Harvesting
• Real-Time Inferencing
• Control Over Sources
• A better way to make data-oriented web sites

Hopefully the term "Data Browser" will come to mean this

Single-Item Page

see a lot information about something

• Person
• Book
• Meeting
• Photograph
• RDF Property

Multi-Item Page

see a little information about each of many items

• The people danbri foaf:knows
• Books on Danny's wish list
• DIG meetings
• Photographs of the W3C team
• Properties with the word "sister" in their docementation

Class-Specific Views

different content and layout for different class items

• People cf friendster, orkut, department pages, okcupid
• Book cf amazon, library card-catalogs
• Meeting cf conference website, agenda-email
• Photograph cf flikr, custom photo sites
• RDF Property cf ontaria 9-board, ...?

Real-Time Harvesting

fetch the data from the web when the user wants it (with caching)

delays in publication are often unacceptable

• Author's feedback-loop
• Meeting agenda changes
• Photographs we just took
• (isn't this obvious)

Real-Time Inference

Let people rely on the formal semantics; let the machines do the work

• subclass/subproperty
• vocabulary mapping rules
• ...?

Control Over Sources

Let readers see (and control) where data came from

one possible design:

• • blue -- on reader's trusted-sites list
  • white -- trusted by uri-owner
  • grey -- 3rd part
  • red -- on reader's highly-suspect list
• combined items (eg inferences) colored at least-trusted level
• mouse-over or nearby [!?] icon for more details
• list of Sources at the bottom of the page, with Mark-As-Trusted and Mark-As-Highly-Suspect
• harvesting and inference depths much greater on more-trusted information

A Better Way To Make Data-Oriented Web Sites?

• What big web-sites are not data oriented?
• Content-Providers:
  • expose RDF views of their database(s)
  • host an off-the-shelf SemWalker installation (or contract for one)
  • can decentralize internally (eg between sales, marketing, shipping, customer relations, etc -- all are automatically integrated on "their" website)

Consumers (Data Users)

• Get better data-oriented web-sites
• Get to use their own data browser
• Incidentally get to query/integrate across provider sites

Market Forces

• Providers looking for best server-side data browsers (SemWalker-clones)
• Consumers looking for best data browsers
• Providers become notably absent from "Data Web" for consumers using their data browsers (presure to expose data)

SemWalker Strategy

• Offers software that lets people build world-class websites just by publishing their public data as RDF
• ... (they might have to add data-organization metadata)
• ... (and an RDF-based transactional-services interface)

Downhill steps to the Semantic Web

Applications

(@@@ expand these examples)

• social networking (view of Person)
• photography (view of a Photo)
• shared vocabularies (view of Ontology, Class, Property)
• shared calendars (view of TimeRange, Event)
• collaboration (view of Project)
• shopping (view of ItemForSale)

@@? Some Requirements

• Allow data source branding
• As usuable as a typical data-oriented website (simple, fast, reliable)
• Customizable views for new Classes
• Privilege of term-origin data (over 3rd party data)
• Quiet-but-available inference
• Good Caching
• Access to search features

The Code

• just my work, so far
• most development driven by Ontaria
  • Key item classes: Ontology, Class, Property
  • Central-hosting plan (portal style), at w3.org
  • Extensive shared indexing - plan for 10^8 triples?

Development History

Harvester

• prioritized threads (hi-queue, low-queue)
• all harvest data visible as RDF
• reads RDF/XML, HTML
• fast, but there's lots of cruft out there
• breaks large graphs into chunks (10K triples)
• can keep old copies of data
• follows redirects, cache-control, etc.

Indexer

• separate thread, running behind harvester
• maps keyword -> (subject, chunk)
• map uri -> chunks that use it
• broken berkeley db implementation

RDF Store

• library('semweb/rdf_db') in memory (an 8-way indexed quad store)
• rdfpage library provides fast loading from disk (pre-parsed)
• pagein_around
  • loads all the chunks that use this URI as subject/value
  • (may need ontology-smarts -- ie pagein_around(rdf:Resource)
• uris given to chunks are all hosted/served
• backward-chaining of rules (as prolog, but incremental depth-limited)

Inference

say more? forward chaining, too, eg for index

• experimentally reads surnia rules for OWL inference

Views

rules for mapping from triples to XHTML trees

• Current implementation makes a tab for each that applies
• I think they're pretty easy/fun to write

The End