Using a Semantic Wiki as a Knowledge Source for Rich Modeling and Question Answering

1
Using a Semantic Wiki as a Knowledge Source for
Rich Modeling and Question Answering

• Vinay K. Chaudhri1, Mark Greaves, Daniel Hansch3,
  Anthony Jameson4, Frederik Pfisterer3,
• Aaron Spaulding1, and Moritz Weiten3
• 1. SRI International, Menlo Park, CA, USA.
• 2. Vulcan Inc, Seattle, WA 3. ontoprise GmbH
• 4. DFKI

2
Symbiosis between KE SW

• Knowledge Engineering
• Expressive knowledge representation
• Sophisticated testing and debugging
• High training requirement
• Semantic Web
• Simpler knowledge representation
• Works by creating links and references
• Almost walk up and use

3
Symbiosis between KE SW

• AURA
• Acquires knowledge for deductive Q/A that can be
  used for answering AP questions in sciences
• Uses a DL style class taxonomy, and logic
  programming style rules with many extensions
• Requires 40 hours of training for knowledge
  formulation
• Semantic Media Wiki
• Tool for online authoring of semantic web content
• Captures knowledge at the level of RDFS
• Almost walk up and use system

4
Symbiosis between KE SW

• Can we use the Semantic Media Wiki to capture
  knowledge that could be used for Q/A in AURA?
• Factual knowledge
• The atomic number for hydrogen is 1
• The solubility constants
• Taxonomic knowledge
• Eukaryotic and Prokaryotic are two types of cells

5
Symbiosis between SW KE Payoffs

• We can make use of contributions from users with
  much less training than 40 hours needed for AURA
• Knowledge creation is faster, distributed, and
  cheaper
• Manage the evolution of concepts and link types
  in the Semantic Wiki
• Collaboration and consensus building ontology tool

6
Outline

• The AURA System
• Media Wiki
• AURA/Media Wiki Symbiosis

7
Outline

• The AURA System
• Media Wiki
• AURA/Media Wiki Symbiosis

8
Introduction

• Long term goal to build a Digital Aristotle an
  application that can answer questions on a
  variety of topics and provide user and domain
  appropriate explanations.
• Mid-term goal to pass Advanced Placement tests
  in Chemistry, Biology, and Physics
• Short-term goal Enable scientists to author
  knowledge and high school students to ask
  questions
• Limited to 50 pages from each domain

9
AURA

• Automated User-Centered Reasoning and
  Acquisition System
• Aura is a tool to help users formalize knowledge
• Aura can then reason with that knowledge
• So users can ask questions and understand the
  answers.

10
AURA Design

• Extensive domain analysis identified four classes
  of frequently occurring textbook knowledge
• Conceptual knowledge
• Equations
• Tables
• Diagrams
• User surveys revealed three classes of user
  requirements
• Blank slate problem
• Support for full life cycle
• Training and usability

11
Textbook Knowledge Types
2Ca(s) O2(g) ? 2CaO(s) h ½gt2
Conceptual
Equations
Diagrams
Tables
12
AURA Desktop
13
Document Rooted Interface
14
Knowledge Capture Conceptual Knowledge

• Concept Maps

15
Knowledge Capture Conceptual Knowledge

• Based on prior work with Shaken CLIB
• Clark et. al., KCAP2001,
• Chaudhri, et. al. EKAW, 2003
• Barker et. al., KCAP2001

• (forall ?c
• (gt (instance-of ?c Eucaryotic-Cell)
• (exists ?x ?y ?z
• (and
• (instance-of ?x Nucleus)
• (instance-of ?y Chromosome)
• (instance-of ?z Plasma-Membrane)
• (has-part ?c ?x) (has-part ?c ?y)
• (has-part ?c ?z) (is-inside ?y ?x)))))

16
Knowledge Capture Equations
17
Knowledge Capture Equations
18
Knowledge Capture Tables
19
Question Formulation Controlled English
An alien measures the height of a cliff by
dropping a boulder from rest and measuring the
time it takes to hit the ground below. The
boulder fell for 23 seconds on a planet with an
acceleration of gravity of 7.9 m/s2. Assuming
constant acceleration and ignoring air
resistance, how high was the cliff?
?
A boulder is dropped. The initial speed of the
boulder is 0 m/s. The duration of the drop is 23
seconds. The acceleration of the drop is 7.9
m/s2. What is the distance of the drop?
20
Question Formulation Interface
21
Explanation Interface
22
Efficacy of knowledge in answering questions
Domain Number of Questions Correct Correct
Domain Number of Questions 2006 2007
Biology 146 38 38
Chemistry 86 38 70
Physics 131 19 71
Knowledge Formulation 6 domain experts, 2 per
domain, 40 hours of AURA training, 80 hours of
knowledge formulation
Question Formulation 6 new domain experts, 2 per
domain, 6 hours of AURA training, 16 hours of
question asking
23
Outline

• The AURA System
• Media Wiki
• AURA/Media Wiki Symbiosis

24
Wikipedia Article on Organelles
25
Source Text of That Article
26
Fact Box Summarizing the Annotations
27
The Query Interface
28
Auto-completion in the Query Interface
29
Table Showing the Result of a Query
30
The Source Text of the Result Table
31
Adding Annotations
32
Ontology Browser
33
Open Source Availability
http//sourceforge.net/projects/halo-extension/
34
Outline

• The AURA System
• Media Wiki
• AURA/Media Wiki Symbiosis

35
Technical Design Issues

• The knowledge may not be clean
• Use a Wiki Gardener who can serve as an editor of
  the online contributions
• Set up negative feedback loops that encourage and
  help users to correct problems as they stumble
  upon them
• Empty row for Centrosome in table ? Visit
  Centrosome page and add or correct annotations
• (Other negative feedback loops not shown in the
  slides)
• Vocabulary mismatches between AURA Wiki
• Use a mapping tool to relate the two vocabularies
• Prime Wiki with the AURA vocabulary

36
Example Use Case

• AURA knowledge formulation engineer searches for
  knowledge during knowledge formulation
• The KFE notices useful information in Wiki
• The KFE maps the knowledge into AURA
• The knowledge is translated into AURA and
  available for querying

37
AURA User Searches for Information
38
AURA User Notices Useful Information
39
AURA user maps knowledge into AURA
40
Knowledge is available for Q/a
41
Status and Challenges

• Giving guidance on what to annotate
• Identify missing information by queries
• Conceptual KE tasks partonomy
• Suggest annotations
• Representation mismatches
• Concept vs. Individual
• A large scale evaluation is planned for Fall 2008