Tiddler: Customised publishing based on XML profiles and XML data sources

1
Tiddler Customised publishing based on XML
profiles and XML data sources

• François Paradis, Cécile Paris,
• Anne-Marie Vercoustre, Stephen Wan, Ross
  Wilkinson, MingFang Wu

CSIRO Mathematical and Information Sciences
2
Outline

• Motivation
• Examples
• Current approaches
• Our approach
• How it works?
• Analysis and Conclusion

3
Motivation Why customised publishing?

• Too much information people want less
  information but more relevant to their need,
  knowledge, or task
• On different devices at different times
  paper, Web, WAP
• (To build customer relationship)

4
Examples

• Customised Travel Guides
• Depending on who (preferences), where to go ,
  when to go,
• Depending on when/where to use it
• Corporate brochures
• Depending on who you are and your current
  interest(s)

5
Current techniques

• Distinct versions of manually crafted documents
  one for printing, one for the Web. No
  personalisation
• Word - HTML Latex - HTML HTML- WAP
• Information Retrieval personalisation through
  queries, synthesis of the results no much
  coherence
• Document generation from database queries and
  different stylesheets coherence but not high
  level semantic of the resulting document. Limited
  type of sources
• Document generation using NL techniques relies
  on the availability of knowledge base in
  appropriate format

6
Tiddler approach

• Exploits both language generation and IR-document
  synthesis approaches
• Coherence preserved
• Wide variety of data sources (including web
  pages) accessible
• Dynamically plan documents
• Customise information using user models
• Generate documents for multiple media types
  (Paper, Palm Pilots, Web browsers, Mobile Phones)

7
System Architecture
Virtual Document Planner
NORFOLK
Content Planner
Discourse Rules
Presentation Planner
Surface Generator
Customised Documents
8
User Model

• Include
• preferences,
• information need,
• Context (device)
• historic
• Collected via a G.U.I. Interface
• Used to
• customise information to user
• determine layout and content detail depending on
  media
• encapsulate some Users Goal
• Goal is about information need
• Virtual Document Planner resolves goal using
  Planning techniques

9
Input User Model

• Name Zoe
• Medium Palm Pilot
• Destination Melbourne
• Date 1 June-15 June 2001
• Activities Cycling, Opera, Major Mitchell
• Travel Information
• Accommodation (backpacker)

10
XML Representation

Zoe
Melbourne
01 June
2001
15 June
2001
Cycling, Opera, Major
Mitchell
Backpackerange
palm-pilot .

11
Output Palm Pilot Version
12
Output Web Version
13
Virtual Document PlannerOverview 1

• The Virtual Document Planner
• uses Planning Techniques
• Goal achieved by finding subgoals that satisfy it
• Subgoals are linked by rhetorical relations
• Subgoals satisfied by
• other decomposable subgoals
• primitive subgoals

14
Virtual Document PlannerOverview 2

• The Virtual Document Planner
• produces a branching tree structure
• Node information need goal
• Nodes in branches subgoals
• Nodes linked by rhetorical relations
• Subgoals and Goals represent
• content selection
• presentation decisions

15
Tree for Zoe Example
Enablement
Preparation
Background
Title, Source
Joint
Further Contact
General Information
Joint
Hotels
Opera
Cycling
Major Mitchell
16
Virtual Document Planner Sub-stages

• Three substages
• The Content Planner
• The Presentation Planner
• The Surface Generator

17
Virtual Document Planner Sub-stage 1

• The Content Planner
• uses Goal Planning
• produces a tree structure
• nodes document content
• Branches rhetorical relations that may be
  realised with discourse markers

18
Virtual Document Planner Sub-stage 2

• The Presentation Planner
• Leaves of the tree chosen content
• Leaves expanded with layout mark-up of document
• Mark-up depends on document organisation
• Customised for particular media type.

19
Virtual Document Planner Sub-stage 3

• The Surface Generator
• Dependent on medium
• Content and layout mark-up are mapped to
• text
• XML
• HTML
• WML
• Natural Language
• graphics
• pictures
• tables
• lists

20
Data Sources

• Norfolk technology
• provides interface between
• Virtual Document Planner
• Data sources
• Data Sources originate from
• corporate data bases
• existing web pages of known layout (wrapping)
• Data Sources can be
• static Norfolk retrieves content in advance -
  XML
• dynamic Norfolk retrieves content as needed by
  Virtual Document Planner

21
Why are Dynamic Documents useful?

• A document can
• be composed using most up-to-date information
• customise information to user
• tailor content to particular query
• tailored to a particular media

22
What are the limitations of current dynamic pages?

• Dynamic pages are often
• statically planned with templates and stylesheets
• Templates grow exponentially in number as
  document becomes more flexible
• represented in program language code
• makes maintenance more difficult
• limited to filtering at document level for
  customisation
• required to maintain separate templates for
  different media

23
Conclusions (1)

• Tiddler Advantages
• Easier to maintain because
• Documents use goal planning, not template based
• Document Rules not in a program language code
• Customisation filters and uses relevant
  information from parts of documents
• Information can be gathered from multiple
  sources
• Documents for different media are generated from
  the same document skeleton
• Only need to update the skeleton

24
Conclusions (2)

• Future Work
• - Reasoning about the discourse to provide
  feedback/explanations
• - Dynamic and complex user model to deal
  with history of information delivery
• - Complex user model to build customer
  relationship