21
Weaving Your Own Semantic Web Dennis Quan December 4, 2002

Weaving Your Own Semantic Web Dennis Quan December 4, 2002

Embed Size (px)

Citation preview

Weaving Your Own Semantic Web

Dennis QuanDecember 4, 2002

“Information at your fingertips” Where are we now? How did we get here? Where are we headed?

Who has done this before? Why might this not work? What can we do about it? What will “the user” think?

The status quo The Web is a great place to find all

sorts of information: Weather forecasts News reports Stock charts Phone numbers and addresses TV program schedules and reviews Airline reservations … and much, much more

The origins of the Web Physicist turned hacker Tim Berners-Lee

developed method for linking together network-accessible documents

HTML: easy to read, easy to write, easy to share

HTTP: universal transport for getting at shared documents via Web browsers

Apache, Perl, et al.: easy to hack together scripts for producing Web content en masse

Mosaic, Internet Explorer, Mozilla, etc.

The future of the Web Today’s Web is great, but…

Users: give them an inch; they want a mile Vicious (virtuous?) cycle of automation

Right now most Web content is human-readable “04 05 2002”: part number? birthdate? price? Possibilities for automation are limited

Screen scraping: pre-interpreting pieces of Web content for use by scripts

Semantic Web: make Web content machine-readable in the first place

Oh, the possibilities! Once content is directly interpretable, barriers to

creative use of such content will be lowered Appointment scheduling Price comparison and negotiation Ontology-based search

Examples: Find me the cheapest French-speaking city to fly to in

March and a hotel others found to be “romantic”. Schedule the meetings that must occur today for the

afternoon and postpone the rest until the next 3 days. Where can I buy A Tale of Two Cities for the

cheapest? I’m willing to buy used if the cost savings is at least 50%.

Lingua franca

Resource Description Framework (RDF): “circle and arrow diagram” method for encoding knowledge

A Tale of Two Cities

Booktype

$10.95

price

Charles Dickensauthor

Agents

Programs that do things on behalf of humans

A Tale of Two Cities

$10.95price

$4.95

used price

Acme Books

says

Honest Joe’s Used Books

says

condition Good

Déjà vu?

Flexible data representation Knowledge representations, ontologies

and descriptive logic systems? Relational databases?

Number crunching and deduction Internet price search engines? Perl scripts? Multi-agent environments?

Problem #1: information Information for the Semantic Web must

come from somewhere CyC approach

Spend $25m and 20 years time Results in highly consistent corpus Problem: requires $25m and 20 years time

Distributed approach Piece by piece incrementally Each user contributes Problem: requires tools for inputting information

Problem #2: “Grandma” Grandma doesn’t know

about SLAD-DOS: Scripting Logic Agent interaction Data types Distributed systems Ontologies Schemas

Technology irrelevant if user interface cannot expose it

Problem #3: the web monkey

Web monkeys like: Simple, easy-to-understand languages

(e.g., JavaScript, Perl, HTML) Granular, hands-on, reusable

components (e.g., CGI scripts, Web pages, Java applets)

Ability to cycle through edit-run-debug quickly (e.g., with a Web browser)

The generation gapGrandma wants to: Tell her friends how

great the toaster she bought is

Find a romantic comedy on TV tonight

Get a doctor’s appointment when Days of Our Lives isn’t on

See latest pictures of grandchildren

Web monkey will need: Distributed, P2P

database with flexible schema

Sophisticated Boolean query language

Online representation of personal calendars and agent negotiation protocol

Content management system

RDF to the rescue Distributed: easily shared

between systems and highly granular

Flexible: doesn’t restrict how people think about their information

Plus all the benefits of 50 years of AI and database research

RDF

Web monkeys like toys Standard RDF databases RDF-enabled scripting

language Distributed agent

communication layer Transports RDF over SOAP, POP3,

SMTP, etc. Drag and drop ontology

designers

… kind of like httpd, perl, and mysql

DB + scripting

RDF

Toys that let users play If users don’t tell their

computers things, agents will have nothing to work with PIM that automagically records

calendar, address book, e-mail, to-do list, etc. as RDF

Editors that can take RDF ontologies written by developers and intelligently allow input from users

… kind of like Web browsers and e-mail clients

UI components

DB + scripting

RDF

Making use of the toys Users must be able to ask their

computers for their information Natural language schemas for

mapping onto RDF representations Natural language query engines

(e.g. START) Agents must be made easily

accessible Users maintain their own agents

much like they do their bookmark collections

… kind of like Google and Priceline.com

Agents + search

UI components

DB + scripting

RDF

Sharing your toys Not all users will understand how to model

data Let those who can share their ontologies Make the UI capable of finding these ontologies

automatically Must also model “hints” or “templates” that give

users suggested defaults UI components and agents must also be

sharable “Onto-Google”?

Client? Server?What’s the difference?

Both users and developers create content

Both clients and servers store information in RDF

Agents can reside on users’ machines (personal agents) or can be distributed across the Internet (like Web Services)

Truly peer-to-peer

A fantasy?

Exercises for the reader My home page

(http://www.ai.mit.edu/people/dquan/) Haystack

(http://haystack.lcs.mit.edu/) Semantic Web

(http://www.w3.org/2001/sw/) RDF

(http://www.w3.org/RDF/)

Thank you for your attention