The Exchange of Retrieval Knowledge about Services between Agents

1

• The Exchange of Retrieval Knowledge about
  Services between Agents

2
Can computer systems make experiences
3
Outline

• Brief introduction of Kühnels assistant agents
• The role of TCBR to improve this agents
• Merging of Case Retrieval Nets
• Implementation issues
• Contribution to Experience Management

4
Kühnels assistant agents

• Personal assistant agents that distribute how-to
  knowledge
• executing services on demand of user (service
  utility program like print a file)
• planning with sub-services if necessary
• collaborating with other agents by
• remote execution
• exchange of services

5
A service description
6
User requests

• How to find appropriate services to a user
  request?
• Navigate? - only for a small service library
• String matching? - bad recall
• Textual Case-Based Reasoning!
• The service descriptions are the cases.
• The retrieval is performed by means of a Case
  Retrieval Net (CRN).

7
Textual CBR

• Case AV pairs textual sections, mapped on
  sets of information entities by means of
  dictionaries, e.g. printprintsprinting
• Similarity function computes the best matching
  cases of the case base concerning a query, uses
  local similarity values, e.g.

8
Retrieval of Texts in a CRN
Opens Yahoo in an Internet Explorer
Case Information Entities Query
__YAHOO__
Starting Yahoo with Netscape
9
Retrieval of Texts in a CRN
Opens Yahoo in an Internet Explorer
Case Information Entities Query
__YAHOO__
Starting Yahoo with Netscape
10
Retrieval of Texts in a CRN
Opens Yahoo in an Internet Explorer
Case Information Entities Query
__BROWSER__
__YAHOO__
__START__
__OPEN__
__IE__
__NETSCAPE__
Starting Yahoo with Netscape
11
Retrieval of Texts in a CRN
Opens Yahoo in an Internet Explorer
Case Information Entities Query
__BROWSER__
__YAHOO__
__START__
__OPEN__
__IE__
__NETSCAPE__
Starting Yahoo with Netscape
12
Retrieval of Texts in a CRN
Opens Yahoo in an Internet Explorer
Case Information Entities Query
__BROWSER__
__YAHOO__
__START__
__OPEN__
__IE__
__NETSCAPE__
Starting Yahoo with Netscape
13
Central vs. personal CRNs

• Central approach in the manner of yellow pages
• Easier to manage, e.g. by a central administrator
• Rather a client-server than an autonomous agent
  model
• Inconsistencies and ambiguities due to different
  sub-domains
• Personal approach with initially delivered and
  individually extendable dictionaries
• Individual service libraries with individual CRNs
• Communication of services and parts of the own CRN

14
Merging Case Retrieval Nets
C1
Merged set of case nodes Merged set of
Information Entities
C1
C2
__IEy__
__IEij__
__IEx__
__IEa__
__IEz__
15
Merging Case Retrieval Nets

• Types of possible syntactic conflicts
• Non-identical duplicates of information
  entitiesthe new IE takes all strings of both
  originals
• Duplicates of similarity relationships with
  different weights choose the maximum degree
• Differently assigned strings, e.g.__print__
  print printer__printer__ printer print device

16
Merging Case Retrieval Nets
new relevance arcs
C1
Case nodes Information Entities new
similarity arcs?
C1
C2
__IEy__
__IEij__
__IEx__
__IEa__
__IEz__
17
Merging Case Retrieval Nets

• New relevance arcs rebuild the CRN
• Missing similarity links, e.g.
• Still an open research issue!

__SEARCH ENGINE__
__ALTAVISTA__
__GOOGLE__
18
Implementation

• Java package TCBR that implements a case-based
  retrieval for texts with a Case Retrieval Net and
  dictionaries
• Java package AgentTCBRShell to extend the
  original description of a service with IEs and
  local similarity relationships
• The prototype is implemented and runs as a Java
  application.

19
System Architecture
20
Contribution to Experience Management

• Agents that exchange how-to knowledge in form of
  services
• Transfer of background knowledge
• Integration of received knowledge with the own
  knowledge repository
• Individual treasury of experiences that
  develops during the agents work history

21
Agents are able to make experiences and to share
them!
22
Questions?
23
Long-Lived CBR Systems

• Lifecycle models and their impact on CBR systems
  regarding usage, structure, and maintenance
• ICCBR-03 Workshop in Trondheim, Norway
• Organizers M. Nick, M. Minor
• Important Dates
• April 9, 2003 submission deadline
• June 24, 2003 workshop
• http//www.iccbr.org/2003/index-ws2.html

24
Applications of TCBR

• ExperienceBook www.informatik.hu-berlin.de/cbr-
  ws/EXP_BOOK
• Support of system administrators and common
  computer users during daily work
• Cases problem description with solution, e.g.
  how to print an ASCII file with UNIX
• Lesson learnt by long-term use knowledge
  management to keep it up to date

25
The Knowledge Creation Spiral of Nonaka/Takeuchi
to
explicit
implicit
Socialisation Externalisation Internal
isation Combination
implicit
from
explicit
26
Experience Management (EM)

• deals with the management of experiential
  knowledge within an organization
• main problems
• How to make experiential knowledge explicit?
• How to find (retrieve) it when it is needed?
• Support of EM processes by tools,
• Integration with the usual business processes

27
State-of-the-art EM Systems

• CBR systems experiential knowledge in cases,
  user-triggered retrieval processes
• Ontology-based systems Ontology hierarchy of
  concepts, user browses through domain ontology
• Communities of practise regular meetings of
  experts, user participates or deals with
  documentations of the meetings

28
EM Systems in Long-Term Use

• EM systems learn
• New experiential knowledge which is stored for
  the users
• Internal knowledge to execute or improve the
  systems core tasks
• During usage, EM systems collect an own treasury
  of experiences!

29
Knowledge Sorts of the Treasury

• Knowledge contents (in CBR the cases)
• Own immanent experiences of the system
• Work history of knowledge pieces
• Indexing and other background knowledge
• Valuating knowledge
• Social knowledge
• Knowledge about users