Table Of ContentLecture Notes in Artificial Intelligence 1612
EditedbyJ.G.CarbonellandJ.Siekmann
Subseries of Lecture Notes in Computer Science
3
Berlin
Heidelberg
NewYork
HongKong
London
Milan
Paris
Tokyo
Ralph Bergmann Klaus-Dieter Althoff
Sean Breen Mehmet Go¨ker
Michel Manago Ralph Trapho¨ner
Stefan Wess
Developing Industrial
Case-Based Reasoning
Applications
The INRECA Methodology
2ndEdition
1 3
Authors
RalphBergmann
UniversityofHildesheim,Germany
E-mail:[email protected]
Klaus-DieterAlthoff
FraunhoferIESE,Kaiserslautern,Germany
E-mail:[email protected]
SeanBreen
CPLtd.,Ballinaclash,Co.Wicklow,Ireland
E-mail:[email protected]
MehmetGo¨ker
Kaidara,LosAltos,CA,USA
E-mail:[email protected]
MichelManago
Kaidara,Paris,France
E-mail:[email protected]
RalphTrapho¨ner
StefanWess
empolisGmbH,Kaiserslautern,Germany
E-mail:{Ralph.Traphoener;Stefan.Wess}@empolis.com
Cataloging-in-PublicationDataappliedfor
AcatalogrecordforthisbookisavailablefromtheLibraryofCongress.
BibliographicinformationpublishedbyDieDeutscheBibliothek
DieDeutscheBibliothekliststhispublicationintheDeutscheNationalbibliografie;
detailedbibliographicdataisavailableintheInternetat<http://dnb.ddb.de>.
CRSubjectClassification(1998):I.2,J.1,J.3
ISSN0302-9743
ISBN3-540-20737-6Springer-VerlagBerlinHeidelbergNewYork
ISBN3-540-66182-4(firstedition)Springer-VerlagBerlinHeidelbergNewYork
Thisworkissubjecttocopyright.Allrightsarereserved,whetherthewholeorpartofthematerialis
concerned,specificallytherightsoftranslation,reprinting,re-useofillustrations,recitation,broadcasting,
reproductiononmicrofilmsorinanyotherway,andstorageindatabanks.Duplicationofthispublication
orpartsthereofispermittedonlyundertheprovisionsoftheGermanCopyrightLawofSeptember9,1965,
initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer-Verlag.Violationsare
liableforprosecutionundertheGermanCopyrightLaw.
Springer-VerlagisapartofSpringerScience+BusinessMedia
springeronline.com
(cid:1)c Springer-VerlagBerlinHeidelberg1999,2003
PrintedinGermany
Typesetting:Camera-readybyauthor,dataconversionbyPTP-Berlin,Protago-TeX-ProductionGmbH
Printedonacid-freepaper SPIN:10977200 06/3142 543210
Preface
Adding CBR Back into the Lexicon:
Today’s CBR Technology Not Only Works,
It Makes Strategic Sense
John Ragsdale, Giga Research Director, Forrester Research
Let me start by stating: case-based reasoning is not evil.
This may come as a surprise to those of us who have been in and around the
world of CRM (customer relationship management) and technical-support
knowledge-base products for a number of years. I personally have been guilty of
associating “case-based reasoning” (CBR) with systems that are hardcoded,
inflexible, difficult and expensive to maintain, and hard to use effectively.
But it is important to separate the concept of CBR from its (supposed)
instantiations. Early knowledge-base products created for technical-support agents
in the late 1980s and early 1990s were sometimes billed as CBR. The first such
systems I encountered took a year to implement in my support center (all custom
work by a systems integrator [SI]), and a week after go-live my agents staged a
walkout. Either the system went, or they did. Talk about low-adoption rates.
Many support managers were burned by systems sold as “CBR,” and I have
encountered quite a few other ex-technical support or helpdesk managers, some of
them now VPs of support or even at the CIO level, who have a visceral reaction to
the term “CBR.” Their reaction is often a surprise to software developers and those
from the more academic side of knowledge management.
So before you begin this truly insightful and promising book about case-based
reasoning and learn how to more effectively access and leverage corporate
knowledge, let’s clear up what CBR is not:
– CBR is not a set of solution documents organized in a hardcoded index “tree.”
– CBR does not force agents or customers to go step by step down a predefined
path through the minutia of choices that never exactly fit the situation or
problem.
– CBR does not mean a single static solution to a single static problem that
requires a programmer to add, change or delete.
Having worked with and for software vendors selling intelligent knowledge-
management products geared toward technical support agents and/or customers
performing Web self-service, I’ve often advised against using the term “CBR” for
VI Preface
fear of backlash from prospects, who like me, may have been burned by a really
bad implementation sold as CBR.
But after reading the first edition of this book, and talking with Michel Manago,
one of the co-authors, about what CBR actually means, I think it is time to revive
the CBR term. It’s also time to stop punishing the entire CBR concept for the sins
of a few early software vendors and greedy SIs. And it’s time to move forward
with learning how to use CBR technology to streamline customer service
operations, cut the operating costs of technical support centers, and offer stellar
web self-service options to customers who have the need, the desire, or just the
plain old moxie to want to figure things out on their own.
The Business Case for Easier Access to Knowledge
In most customer service centers, customer interactions are increasing as much as
30 percent year after year, counting all of the new “e” interaction channels, such as
e-mail and chat, in addition to phone calls. The percentage of interactions by
channel is changing as well, as more customers adopt electronic channels. While
the volume of phone calls received by the average support operation may remain
the same, or increase slightly as other interaction channels grow in popularity,
phone calls constitute a smaller percentage of overall interactions. E-mails are
expected to pass phone calls in volume in most industries by 2010.
All of this is forcing companies to fund initiatives for customer self-service,
hoping to entice more and more customers to attempt self-service before calling for
assistance or at least to streamline agent processing of customer questions with
robust knowledge-base options. Ideally, these scalable solutions allow companies
to compensate for increases in interaction volume year after year by minimizing
both the number of interactions involving agents and the time agents spend
processing each interaction.
Expecting even a trained agent (let alone a novice customer) to find useful
information in a static knowledge base by using only a full text search is
unrealistic. Customers cannot possibly know all of the key “symptom” words to
include, and the more textual clues the customer provides in the search string, the
larger the number of possible matches returned. Most customers attempting self-
service will abandon all hopes of resolving their own problems when the text
search returns over 1,000 possible matches. And such results train customers to
never attempt self-service again, forcing them to migrate to the most expensive
support channel: a live phone call with an agent.
Case-based reasoning makes sense in those industries where internal or external
customer support involves disparate or complex hardware or software
configurations, multiple technical environments or platforms, and varying levels of
service or warranty options, as well as vastly different experience levels across the
enterprise or customer base.
Adding CBR Back into the Lexicon VII
Why CBR Makes Sense Now — Technically
CBR solutions have credibility today because three main technological areas have
come a long way since the 1980s: linguistic analysis, consolidated customer
databases, and intelligent analytics.
Natural Language Searching (NLS) comes in many flavors, far too many to
describe and differentiate here. But one area of NLS, linguistic analysis, has moved
out of the laboratory and into packaged software products, allowing a system to
intelligently understand what a customer (or agent) is asking, not just to perform a
keyword search. It’s also about attribution, i.e., the amount of ancillary data that
can be factored into a problem diagnostic session to more succinctly identify the
correct answer or knowledge “chunk.” Before the days of XML and SOAP
interfaces, only minimal data was passed between systems, and few environmental
variables were stored in a single database.
CRM has changed all that. A single CRM customer-information model can
create a “360-degree view” of the customer, listing every product the customer has
purchased, the technical environment in place (Unix vs. Linux vs. NT, or Oracle
vs. Sybase vs. SQL), every problem the customer has experienced in the past, and
any other information a company may choose to capture, such as the user’s skill or
training level, preferred language, etc.
With this amount of attributable information at the disposal of a CBR
application, customers (or agents) need not manually supply background
information (or follow an index tree answering basic questions). The context of
each question is automatically established, allowing a CBR system to more
granularly identify a case that meets the exact question in relation to all
environmental variables. The more the system is used, and the more users indicate
which solutions solved which problems, the more environmental data from all of
those users can be captured within the case framework, thus allowing the system to
continually “learn.” Add to this the advent of analytic engines that can make
intelligent, real-time and even personalized decisions and recommendations based
on any amount of available data, past history, and any rules in place. Though both
software vendors and IT developers are wary of the term “artificial intelligence”
(AI) — often because they were burned by early CBR systems that were said to
use AI but failed to deliver — real-time analytics is here, and vendors have
customer references to prove it.
A lot of things have changed since the mid—1980s. Giving agents tools to
increase productivity and offering robust customer self-service systems is no
longer an academic exercise; it is the only way large support operations can even
come close to meeting customer expectations in this age of budget cuts and staff
downsizing. And today’s CBR-based products don’t take a year to implement; they
can guarantee a positive return on investment in as little as three months. For many
support environments, CBR is a logical choice.
Table of Contents
About This Book...............................................................................................1
Acknowledgments.............................................................................................5
Part I : Smarter Business with Case-Based
Decision Support
.......................................................................7
1. Making Smarter Business Decisions in Less Time.......................9
1.1 Why Case-Based Decision Support?......................................................10
1.2 What Is Case-Based Decision Support?..................................................11
1.2.1 Case-Based Reasoning.................................................................11
1.2.2 Knowledge Discovery and Data Mining......................................13
1.3 Successful Applications of Case-Based Decision Support.....................14
1.3.1 Management of Industrial Knowledge.........................................14
1.3.2 Sales Support...............................................................................14
1.3.3 Reliability Analysis......................................................................14
1.3.4 Technical Support........................................................................15
1.4 Benefits of Case-Based Decision Support..............................................15
1.5 Case-Based Decision Support and Alternative Techniques....................16
1.5.1 Statistics.......................................................................................16
1.5.2 Expert Systems............................................................................16
1.5.3 Databases.....................................................................................17
1.6 Development Process and Case Collection.............................................17
1.7 How Are Cases Retrieved?.....................................................................18
1.8 How Are Cases Represented?.................................................................19
2. Case-Based Reasoning Approaches..................................................21
2.1 The Textual CBR Approach...................................................................22
2.2 The Conversational CBR Approach........................................................23
2.3 The Structural CBR Approach................................................................25
2.4 Comparing the Different Approaches.....................................................28
2.5 Effort Required for the Different Approaches........................................31
2.6 Overall Assessment of the Different CBR Approaches..........................33
X Table of Contents
3. Selected Applications of the Structural Case-Based
Reasoning Approach...............................................................................35
3.1 HOMER: A Case-Based CAD/CAM Help-Desk Support Tool..............36
3.1.1 DaimlerChrysler’s Motivation.....................................................36
3.1.2 Initial Situation at the CAD/CAM Help-Desk.............................36
3.1.3 Project Goals and Benefits...........................................................38
3.1.4 Knowledge Representation..........................................................39
3.1.5 The HOMER System...................................................................40
3.1.6 Users and Roles...........................................................................44
3.1.7 Lessons Learned..........................................................................44
3.1.8 Results Achieved.........................................................................45
3.1.9 Summary......................................................................................46
3.2 A Case-Based Product Catalog of Operational Amplifiers.....................47
3.2.1 The Application Development Process........................................48
3.2.2 Benefit Analysis...........................................................................48
3.2.3 Similarity Functions Used in the Application..............................49
3.2.4 How the Software Works.............................................................50
3.2.5 Status of the Application.............................................................54
3.3 Case-Based Maintenance Support for TGV Trains.................................55
3.3.1 TGV’s Environment....................................................................55
3.3.2 A Hierarchy of Data Models and Case Bases..............................55
3.3.3 Technical Approach for Diagnosis..............................................57
3.3.4 Example of the System in Action................................................58
3.3.5 Integration into the End-User Environment.................................59
3.3.6 Status of the Application.............................................................61
3.4 The Fraunhofer IESE Experience Factory..............................................61
3.4.1 Fraunhofer IESE’s Motivation.....................................................61
3.4.2 Initial Situation............................................................................62
3.4.3 Sharing Project Experience..........................................................63
3.4.4 Benefits of Sharing Explicitly Stored Experience.......................64
3.4.5 Organizing a Best-Practice Repository........................................64
3.4.6 Populating the Case Base.............................................................66
3.4.7 Envisioned Usage of Lessons Learned........................................68
3.4.8 Conclusion...................................................................................70
Table of Contents XI
Part II: Developing Case-Based Application with
the INRECA Methodology
........................................71
4. Practical Guidelines for Developing Case-Based
Reasoning Applications.........................................................................73
4.1 Define Your Measures of Success..........................................................74
4.2 Orient Your Customers...........................................................................75
4.3 Get the Big Picture..................................................................................76
4.4 Think about Maintenance........................................................................77
4.5 Welcome Feedback and Testing.............................................................78
4.6 Measure the Success...............................................................................79
4.7 Direct Access for the Customer..............................................................80
4.8 Market Your Success..............................................................................81
4.9 Expand the Scope....................................................................................82
5. Professional Case-Based Reasoning Application
Development...............................................................................................85
5.1 Why Do You Need the INRECA Methodology?....................................86
5.1.1 How You Benefit from a Methodology.......................................86
5.1.2 What the INRECA Methodology Can Do for You......................86
5.1.3 Ensuring High-Quality Software.................................................87
5.2 Foundations from Software Engineering................................................87
5.2.1 The Experience Factory and the Quality Improvement
Paradigm.................................................................................................88
5.2.2 Software Process Modeling.........................................................90
5.3. The CBR Experience Base......................................................................92
5.3.1 Experience Captured in Software Process Models......................93
5.3.2 Experience on Different Levels of Abstraction...........................93
5.4 How to Apply the INRECA Methodology to New Projects...................94
5.5 Capturing Your Own CBR Experience...................................................96
5.6 The INRECA Methodology and Quality Standards................................98
6. Documenting Case-Based Reasoning Development
Experience...................................................................................................99
6.1 Into the Details of Processes, Products, and Methods...........................100
6.1.1 Software Processes: The Basic Activities..................................100
6.1.2 Technical, Organizational, and Managerial Processes...............101
6.1.3 How Processes Interact..............................................................101
6.1.4 Combining Processes to Process Models...................................103
6.2 Generic and Specific Descriptions........................................................103
6.2.1 Generic Descriptions.................................................................104
