24
Reasoning Framework for Knowledge Management Applications German Workshop on CBR March 15, 2001 Rosina Weber David W. Aha, Nabil Sandhu, Héctor Muñoz-Avila Decision Aids Group Center for Applied Research in Artificial Intellig Naval Research Laboratory University of Wyoming

A Textual Case-Based Reasoning Framework for Knowledge Management Applications German Workshop on CBRMarch 15, 2001 Rosina Weber David W. Aha, Nabil Sandhu,

Embed Size (px)

Citation preview

A Textual Case-Based Reasoning Framework for Knowledge Management

Applications

German Workshop on CBR March 15, 2001German Workshop on CBR March 15, 2001

Rosina WeberDavid W. Aha, Nabil Sandhu, Héctor Muñoz-

Avila

Rosina WeberDavid W. Aha, Nabil Sandhu, Héctor Muñoz-

Avila

Decision Aids GroupNavy Center for Applied Research in Artificial Intelligence

Naval Research LaboratoryUniversity of Wyoming

Decision Aids GroupNavy Center for Applied Research in Artificial Intelligence

Naval Research LaboratoryUniversity of Wyoming

Outline

Introduction Knowledge Management Systems Knowledge artifacts Lessons Learned Systems

Motivation Methodology

NEO domain Case Representation Elicitation Tool Extraction Tool Monitored Distribution Domain Ontology

Problems vs. Solutions Next Steps

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

3 R.Weber, NCARAI-NRL, U. of Wyoming

Knowledge Management Systems

KMS manipulate knowledge to...

….storing, distribute, collect, validate, apply, create, sharing & leveraging knowledge

CORPORATE MEMORY

DOCUMENTS

KNOWLEDGE ARTIFACTS

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

4 R.Weber, NCARAI-NRL, U. of Wyoming

Knowledge artifacts

are structured formalisms that imply essential elements of knowledge for reuse (e.g., when to reuse, what to reuse) well understood and accepted

lessons learned alerts best practices incident reports

Alert systemsLessons Learned systems: our current focus

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

5 R.Weber, NCARAI-NRL, U. of Wyoming

why? what was the originating event success/failure/advice cause

when to reuse? task/contextual info about the process main index guiding distribution

Lessons refer to one

task/activity/decision of a process

originate from successes, failures, or advice

teach something about a work practice that has the potential to generate a positive impact in the targeted process when reused

what to reuse?

what to repeat or avoid

under which conditions?

what is required for the lesson to be applicable?

reuse components

indexing

solution

Weber et al., 2001Intelligent lessons learned systems. International Journal of

Expert Systems Research & Applications, Vol. 20, No. 1, Jan 2001, 17-34.

Weber et al., 2001Intelligent lessons learned systems. International Journal of

Expert Systems Research & Applications, Vol. 20, No. 1, Jan 2001, 17-34.

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

6 R.Weber, NCARAI-NRL, U. of Wyoming

Motivation (i) LLS are not used

lessons are distributed outside the context of reuse

lessons are collected in textual descriptions, so they are:poorly representeddifficult to be retrieved& difficult to be interpreted

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

7 R.Weber, NCARAI-NRL, U. of Wyoming

Motivation (ii)

in terms of reuse elements

artifacts disseminated in the context of external

distribution systems (DS)

reusing knowledge artifacts

share knowledge

Knowledge artifacts as cases

text extraction toolelicitation tool

Domain Ontology + Subset of NLChallenge:Natural language

human users text

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

8 R.Weber, NCARAI-NRL, U. of Wyoming

TCBR Methodology for Knowledge Management Systems

that manipulate knowledge artifacts

Elicitation Tool Extraction tool Case Representation Monitored Distribution Domain Ontology

case base

extraction tool

human users

textualdocuments

format of external distribution

system

artifacts in the format of ext distribution system

domain specific ontology

elicitation tool

Why CBR?

Why textual?

.

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

9 R.Weber, NCARAI-NRL, U. of Wyoming

Noncombatant Evacuation Operations-NEO: military operations to evacuate noncombatants whose lives are in danger to a safe haven

AssemblyPoint

HQ

ISB

safe haven

Noncombatant Evacuation Operations:

military operations to evacuate noncombatants whose lives are in danger to a safe haven

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

10 R.Weber, NCARAI-NRL, U. of Wyoming

Noncombatant Evacuation Operations (NEO)

Assembly Point

NEO site safe haven

ISB

HQ

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

11 R.Weber, NCARAI-NRL, U. of Wyoming

Case Representation Elicitation Tool Extraction tool Monitored Distribution Domain Ontology

TCBR Methodology for Knowledge Management Systems

that manipulate knowledge artifacts

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

12 R.Weber, NCARAI-NRL, U. of Wyoming

Case Representation

Example:1. When/Where to reuse (which task): Registering evacuees

Context/Process: NEO operation2. Under which conditions: The weather is hot and humid. The location is a tropical country.3. What to reuse: Make sure to avoid registration in 3 steps.4. Why (originating event): We implemented registration in 3 steps.

Success/Failure: It was a failure.

Why? It was very time consuming.It caused evacuee discomfort.

Additional elements provided by the domain ontology.

Example:1. When/Where to reuse (which task): Registering evacuees

Context/Process: NEO operation2. Under which conditions: The weather is hot and humid. The location is a tropical country.3. What to reuse: Make sure to avoid registration in 3 steps.4. Why (originating event): We implemented registration in 3 steps.

Success/Failure: It was a failure.

Why? It was very time consuming.It caused evacuee discomfort.

Additional elements provided by the domain ontology.

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

13 R.Weber, NCARAI-NRL, U. of Wyoming

Requirements:

Indentify the audience style Identify reuse & retrieve

components: knowledge, process, conditions of applicability, explanation

Identify the format of components Identify relationships

Requirements:

Indentify the audience style Identify reuse & retrieve

components: knowledge, process, conditions of applicability, explanation

Identify the format of components Identify relationships

Case Representation

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

14 R.Weber, NCARAI-NRL, U. of Wyoming

Elicitation Tool

What: The lesson elicitation tool LET guides and

educates users to submit lessons in the CR It orients with examples and reduces the

amount of text to type by giving drop-down lists to select from

It requests confirmations to orient the user to rethink the experience to be communicated

A domain ontology supports disambiguation at run-time (do not store unless relevant)

Uses a subset of NL based on the CRF by using a template-based elicitation with pre-defined grammar structures to overcome NLP problems

What: The lesson elicitation tool LET guides and

educates users to submit lessons in the CR It orients with examples and reduces the

amount of text to type by giving drop-down lists to select from

It requests confirmations to orient the user to rethink the experience to be communicated

A domain ontology supports disambiguation at run-time (do not store unless relevant)

Uses a subset of NL based on the CRF by using a template-based elicitation with pre-defined grammar structures to overcome NLP problems

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

15 R.Weber, NCARAI-NRL, U. of Wyoming

Elicitation Tool

Requirements: in connectivity with the domain

ontology be supported by lexicons of

expressions, domain entities and verbs

support conversation to acquire new concepts for the ontology

Requirements: in connectivity with the domain

ontology be supported by lexicons of

expressions, domain entities and verbs

support conversation to acquire new concepts for the ontology

Example:

Example: Microsoft PowerPoint

Presentation

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

16 R.Weber, NCARAI-NRL, U. of Wyoming

What: converts texts into knowledge

artifacts template mining

variant of Information Extraction search for specific descriptions in

selected excerpts of text (structure)

avoids NLP techniques uses methods that contain

knowledge of where to search and what to extract

What: converts texts into knowledge

artifacts template mining

variant of Information Extraction search for specific descriptions in

selected excerpts of text (structure)

avoids NLP techniques uses methods that contain

knowledge of where to search and what to extract

Extraction Tool

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

17 R.Weber, NCARAI-NRL, U. of Wyoming

Requirements:

Source text must follow stereotypical style

Source text must have some structure that allows identification of a rhetorical structure

Domain of source text is known

Requirements:

Source text must follow stereotypical style

Source text must have some structure that allows identification of a rhetorical structure

Domain of source text is known

Extraction Tool

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

18 R.Weber, NCARAI-NRL, U. of Wyoming

Example: Method converting textual lessons into the case representation framework:

“In field recommended action, search for expressions such as (in this order): make sure , ensure, should. When (if) one of these is found, extract content beginning right after the expression found until the next period.”

Example: Method converting textual lessons into the case representation framework:

“In field recommended action, search for expressions such as (in this order): make sure , ensure, should. When (if) one of these is found, extract content beginning right after the expression found until the next period.”

Extraction Tool

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

19 R.Weber, NCARAI-NRL, U. of Wyoming

Monitored Distribution

What: a framework to solve the lesson

distribution gap disseminate knowledge in the context

of targeted processes (just in time knowledge)

infrequent variable experiential knowledge

allows executable implementation of knowledge

What: a framework to solve the lesson

distribution gap disseminate knowledge in the context

of targeted processes (just in time knowledge)

infrequent variable experiential knowledge

allows executable implementation of knowledgeExample: Example: Microsoft PowerPoint

Presentation

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

20 R.Weber, NCARAI-NRL, U. of Wyoming

Monitored DistributionRequirements: The conversion of the knowledge artifacts into the format of the external distribution systems.

Requirements: The conversion of the knowledge artifacts into the format of the external distribution systems.

Evaluation:We have evaluated the monitored distribution in two domains:

Evaluation:We have evaluated the monitored distribution in two domains:

domain/measure

travel duration

NEO duration

NEO casualties

no lessons

9h45

39h50

11.5

with lessons reduction

32h48

8.7

9h14 5%

18%

24%

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

21 R.Weber, NCARAI-NRL, U. of Wyoming

Domain Ontology

What: A hierarchical model of domain

knowledge where concepts are organized according to their commonalities and meaning

It supports the CR, similarity assessment, knowledge elicitation, text extraction, and the conversion of artifacts into the format of external distribution systems

We are currently investigating corpus analysis to learn lexicons, concepts, and relations from about 40,000 lessons

What: A hierarchical model of domain

knowledge where concepts are organized according to their commonalities and meaning

It supports the CR, similarity assessment, knowledge elicitation, text extraction, and the conversion of artifacts into the format of external distribution systems

We are currently investigating corpus analysis to learn lexicons, concepts, and relations from about 40,000 lessons

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

22 R.Weber, NCARAI-NRL, U. of Wyoming

Domain OntologyExample: Condition complement:

“ it is a disaster relief operation.” Operation cause:

“ disaster relief” Operation hostility level:

“permissive” to the “hostility level”.

Example: Condition complement:

“ it is a disaster relief operation.” Operation cause:

“ disaster relief” Operation hostility level:

“permissive” to the “hostility level”. Requirements: Knowledge acquisition from domain experts Automatic acquisition

Requirements: Knowledge acquisition from domain experts Automatic acquisition

Microsoft PowerPoint

Presentation

Introduction Knowledge

Management Systems

Knowledge artifacts

Lessons/ Learned Systems

MotivationMethodology

NEO domain Case

Representation Elicitation Tool Extraction Tool Monitored

Distribution Domain

Ontology

Problems vs. Solutions

Next Steps

23 R.Weber, NCARAI-NRL, U. of Wyoming

Next Steps

learning ontology support conversation to

acquire new concepts for the ontology

evaluating the elicitation tool implementing text extraction

for all reuse components evaluating extraction tool

Fourth International Conference on CBR

30 July – 2 August 2001Vancouver, BC (Canada)Premiere CBR meetingIndustry DayExhibition5 WorkshopsGreat social schedule!

www.iccbr.org/iccbr01www.iccbr.org/iccbr01

Chair: Qiang Yang Program Chairs: David W. Aha, Ian WatsonWorkshop Chairs

Rosina Weber & Cristiane Gresse von Wangenheim

Chair: Qiang Yang Program Chairs: David W. Aha, Ian WatsonWorkshop Chairs

Rosina Weber & Cristiane Gresse von Wangenheim

1. Process-Oriented KM Kurt D. Fenstermacher, Carsten Tautz2. Soft Computing Simon C.K. Shiu

3. Authoring Support David Patterson, Agnar Aamodt, Barry Smyth 4. Creative SystemsCarlos Bento, Amilcar Cardoso

 5. CBR in E-Commerce Robin Burke

1. Process-Oriented KM Kurt D. Fenstermacher, Carsten Tautz2. Soft Computing Simon C.K. Shiu

3. Authoring Support David Patterson, Agnar Aamodt, Barry Smyth 4. Creative SystemsCarlos Bento, Amilcar Cardoso

 5. CBR in E-Commerce Robin Burke