chapter 3 informal introduction to similarity-based and case-based reasoning stand 20.12.00

Chapter 3

Informal Introduction to Similarity-Based and

Case-Based Reasoning

Stand 20.12.00

- 2 - (c) 2000 Dr. Ralph Bergmann and Prof. Dr. Michael M. Richter, Universität Kaiserslautern

Recommended References

• This lecture is not intended to provide a complete introduction into case based reasoning. It will rather deal only with those aspects which are used in applications to e-c. For additional readings we recommend:– A. Aamodt, E. Plaza: Case-Based Reasoning: Foundational

Issues, Methodological Variations, and System Approaches. AI Communications 7(1) (1994), S.39-59.

– M. Lenz, B. Bartsch-Spörl, H.-D. Burkhard, S. Wess (eds.): Case-Based Reasoning Technology. Springer Lecture Notes in AI 1400, 1998.

– R. Bergmann, S. Breen, M. Göker, M. Manago, S. Wess: Developing Industrial Case-Based Reasoning Application - The INRECA- Methodology. Springer Lecture Notes in AI 1612, 1999.


Case-Based Reasoning (CBR)

• Case-based reasoning (CBR) is a certain technique which was based on analogical reasoning.

• The main intention is to reuse previous experiences for actual problems.

• The difficulty arises when the actual situation is not identical to the previous one: There is an inexactness involved.

• Its main aspect is that CBR-techniques allow inexact (approximate) reasoning in a controlled manner.

• Here we will shortly describe its main features.• Major applications have been fault diagnosis and help

desk systems.


Similarity Based Reasoning

• The central notion in CBR is the concept of similarity. • The methods in CBR have been extended in a way which

allows applications to other problems rather than reusing previous experiences:– in electronic commerce e.g. to product selection.

• This is due to an abstract formulation of the similarity concept.

• In particular, the main algorithms of CBR can still be applied to these new situations.

• We will first describe the original technique informally and then proceed to the extensions.


What is CBR?

• Case-based reasoning is [...] reasoning by remembering.Leake, 1996

• A case-based reasoner solves new problems by adapting solutions that were used to solve old problems.Riesbeck & Schank, 1989

• Case-based reasoning is a recent approach to problem solving and learning [...] Aamodt & Plaza, 1994

• Case-based reasoning is both [...] the ways people use cases to solve problems and the ways we can make machines use them.Kolodner, 1993


History of CBR in USA

• Roger Schank, Yale University: Cognitive Science– 1977: Scripts for knowledge representation (Schank, Abelson)

– 1983: Dynamic Memory Theory, Memory Organization Packets CYRUS: First implemented CBR-System

(Kolodner)– 1983-1988: Other Systems, e.g., JUDGE, SWALE, CHEF

• Bruce Porter, Austin Texas: Concept Learning

– 1986-89: System PROTOS (Exemplar-based concept representation)

• Edwina Rissland, U. of Massachusetts: Cases in Law (since 1983)

– 1990-92: Systems HYPO (Ashley) and CABARET (Skalak)

• Jaime Carbonell & Manuela Veloso, Carnegie Mellon U.: Analogy– since 1990 Prodigy/Analogy: Case-based Planning using analogy

Interest in CBR is increasing in USA (new research groups), since 1988 several DARPA and AAAI Workshops


History of CBR in Europe

• Michael M. Richter, U. Kaiserslautern, Germany: CBR for Expert Systems– 1988-1991 Systems MOLTKE and PATDEX (technical diagnosis)– since 1991 Case-Based Planning: Systems Caplan/CbC, PARIS – since 1992 European Projects INRECA, INRECA-II, WEBSELL

• Ramon Mantaras, Enric Plaza, IIIA Blanes, Spain: CBR and ML– 1990 Case-Based Learning for medical diagnosis

• Agnar Aamodt, U. Trondheim, Norway: CBR and Knowledge Acquisition– 1991 System CREEK: Integration of Cases and general knowledge

• Mark Keane, Trinity College, Dublin: Cognitive Science– since 1988 Theory of analogical reasoning

Since 1991 Increasing interest in Europe (several new research groups)– 1991 First German CBR Workshop (AKCBR, GWCBR)– 1993 First European CBR Workshop (EWCBR)– 1995 First International CBR Conference (ICCBR)


Extensions to E-Commerce• The extensions to applications in electronic commerce have

been developed in various research projects like

• WEBSELL (Esprit project, No. 27068 ):– tec:inno GmbH, Germany (prime contractor)– Interactive Multimedia Systems, IMS, Ireland – IWT Magazin Verlags GmbH, Germany– Adwired GmbH, Switzerland– Trinity College Dublin, Ireland– University of Kaiserslautern, Germany

• READEE (Rhineland-Palatinate):– tec;inno GmbH, Germany– Engineering Office „Conradi & Partner“– University of Kaiserslautern, Germany


Example Applications (Overview)

• Commercial applications are, e.g.:• Accommodation search in the Müritz region:

http://www.mueritz.de/• Search for Operational Amplifiers (Analog Devices):

http://imsgrp.com/analog/psearch.htm• Support for networking products (3Com):

http://knowledgebase.3com.com/• Travel Agency - Last Minute Trips (Check Out Touristik):

http://www.reiseboerse.com/


Case-Based Reasoning (CBR)

• Basic Ideas:– Store previous experience (case)– Solve new Problems by selecting and reusing cases– Store new experience again

• Replaces 0-1-logic by approximation• Is a well-founded technology:

– Mathematically– Algorithmically– With respect to software technology– Supported by experiments and applications– Business success


What is a Case ?

• A case has two parts:– Description of a problem or a set of problems

(generalized case)– Description of the solution of this problem

(formally or informally)• Possibly additions like explanations, comments on the

quality of the solution etc.

• Cases represent experiences : They record how a problem was solved in the past


Different Case Representations

Free text: textual CBR (tecInno’s CBR answers, ServiceSoft, Serviceware, Inference’s casepoint 1st step)

Lists of questions and answers: conversational CBR (Halley enterprise, Inference’s casepoint 2nd step). No common case structure.

Database like representation: structural CBR (tecInno’s CBR Works and orenge, Acknosoft’s KATE, CaseBank, Isoft’s Recall, CSI’s Remind)


Structured Case Representation• Many different case representations are used (see chapters 4 and

5):– Depend on requirements of domain and task– Structure of already available case data

• Flat feature-value list– Simple case structure is sometimes sufficient for problem solving– Easy to store and retrieve in a CBR system

• Object-oriented representations– Case: collection of objects (instances of classes) – Required for complex and structured objects

• For special tasks:– Graph representations: case = set of nodes and arcs – Plans: case = (partially) ordered set of actions– Predicate logic: case = set of atomic formulas


How to Use a Case

Solution ?

Solution adaptation

ProblemProblemof the case

Solutionof the case

In general, there is no guarantee for getting good solutionsbecause the case may be „too far away“ from the problem. Therefore the problem arises how to define when a case is „close enough“.


How to Use a Case-Base

• A case base is a data base of cases• If a new problem arises one will use a case from the

case base in order to solve the problem• If we have many cases then the chance is higher to

find one with a suitable solution• Because the given problem is usually not exactly in

the base one wants to retrieve a case which solved a problem which is „similar enough to be useful“

• Hence, the notion of similarity is central to CBR• The concept of similarity based retrieval is compared

with data base retrieval


Knowledge Container of CBR

Solutiontrans-

formation

Casebase

VocabulaySimilaritymeasure

DataInformationKnowledge

StorageCompilation

Cases have notto be understood inorder to be stored

In order to solve problems one needs knowledge. Where is it located: In knowledge containers.

A task of knowledge management is the maintenance of the containers.


Typical Problems Handled with CBR: Classification and Diagnosis

• A class is a certain subset of some universe and a classification assigns to each element one or more classes to which it belongs.

• In fault diagnosis the classification is only the first step:

Observations diagnosis repair

classification Domain rules

Diagnosis occurs frequently in After Sales Support, see chapter 12


A Simple Example (I) OverviewTypical Scenario: Call Center

• Technical Diagnosis of Car Faults:– symptoms are observed (e.g., engine doesn’t start) and

values are measured (e.g., battery voltage = 6.3V)

– goal: Find the cause for the failure (e.g., battery empty) anda repair strategy (e.g., charge battery)

• Case-Based Diagnosis:– a case describes a diagnostic situation and contains:

• description of the symptoms

• description of the failure and the cause

• description of a repair strategy

– store a collection of cases in a case base

– find case similar to current problem and reuse repair strategy


A Simple Example (II)What does a Case Look Like?

• A case describes one particular diagnostic situation• A case records several features and their specific values

occurred in that situation

A case is not a ( general) rule !!

Feature Value

Problem (Symptoms)• Problem: Front light doesn’t work• Car: VW Golf IV, 1.6 l• Year: 1998• Battery voltage: 13,6 V• State of lights: OK• State of light switch: OK

Solution• Diagnosis: Front light fuse defect• Repair: Replace front light fuse

CASE

1


A Simple Example (III)A Case Base With Two Cases

• Each case describes one particular situation

• All cases are independent of each other

Problem (Symptoms)• Problem: Front light doesn’t work• Car: VW Golf III, 1.6 l• Year: 1996• Battery voltage: 13,6 V• State of lights: OK• State of light switch: OK


CASE

1

Problem (Symptoms)• Problem: Front light doesn’t work• Car: Audi A4• Year: 1997• Battery voltage: 12,9 V• State of lights: surface damaged• State of light switch: OK

Solution• Diagnosis: Bulb defect• Repair: Replace front light

CASE

2


A Simple Example (IV)Solving a New Diagnostic Problem

• A new problem has to be solved• We make several observations in the current situation• Observations define a new problem• Not all feature values have to be known

Note: The new problem is a “case” without solution part

FeatureValue

Problem (Symptom):• Problem: Break light doesn’t work• Car: Audi 80• Year: 1989• Battery voltage: 12.6 V• State of light: OK


A Simple Example (V)

• When are two cases similar?• How to rank the cases according to their similarity?

Similarity is the most important concept in CBR !!

• We can assess similarity based on the similarity of each feature• Similarity of each feature depends on the feature value.• BUT: Importance of different features may be different

New ProblemNew Problem

CASE

x

Similar?Similar?

Compare the New Problem with Each Case and Select the Most Similar Case :


A Simple Example (VI)Similarity Computation

• Assignment of similarities for features values.

• Express degree of similarity by a real number between 0 and 1

• Examples: – Feature: Problem

– Feature: Battery voltage (similarity depends on the difference)

• Different features have different importance (weights)!– High importance: Problem, Battery voltage, State of light, ...

– Low importance: Car, Year, ...

Not similar Very similar

Front light doesn’t work Break light doesn’t work

Front light doesn’t work Engine doesn’t start

0.8

0.4

12.6 V 13.6 V

12.6 V 6.7 V

0.9

0.1


A Simple Example (VII)Compare Similarity and Case 1

• Similarity computation by weighted average similarity(new,case 1) = 1/20 * [ 6*0.8 + 1*0.4 + 1*0.6 + 6*0.9 + 6* 1.0 ] = 0.86

Problem (Symptom)• Problem: Break light doesn’t work• Car: Audi 80• Year: 1989• Battery voltage: 12.6 V• State of lights: OK

Problem (Symptoms)• Problem: Front light doesn’t work• Car: VW Golf III, 1.6 l• Year: 1996• Battery voltage: 13.6 V• State of lights: OK• State of light switch: OK


0.80.40.60.91.0

Very important feature: weight = 6

Less important feature: weight = 1


A Simple Example (VIII)Compare Similarity and Case 2

• Similarity computation by weighted average similarity(new,case 2) = 1/20 * [ 6*0.8 + 1*0.8 + 1*0.4 + 6*0.95 + 6*0 ] = 0.585

Case 1 is more similar: due to feature “State of lights”

Problem (Symptom)• Problem: Break light doesn’t work• Car: Audi 80• Year: 1989• Battery voltage: 12.6 V• State of lights: OK

Problem (Symptoms)• Problem: Front light doesn’t work• Car: Audi A4• Year: 1997• Battery voltage: 12.9 V• State of lights: surface damaged• State of light switch: OK


0.80.80.40.950

Very important feature: weight = 6

Less important feature: weight = 1


A Simple Example (IX)Reuse the Solution of Case 1

• New Solution:• Diagnosis: Break light fuse defect• Repair: Replace break light fuse

Problem (Symptom):• Problem: Break light doesn’t work• Car: Audi 80• Year: 1989• Battery voltage: 12,6 V• State of light: OK

Adapt Solution:

How do differences in the problem affect the solution?

Problem (Symptoms):

• Problem: Front light doesn’t work

• ...

Solution:

• Diagnosis: Front light fuse defect

• Repair: Replace front light fuse

CASE

1


A Simple Example (X)Store the New Experience

CASE

3

Problem (Symptoms):• Problem: Break light doesn’t work• Car: Audi 80• Year: 1989• Battery voltage: 12.6 V• State of lights: OK• State of light switch: OK

Solution:• Diagnosis: break light fuse defect• Repair: replace break light fuse

If diagnosis is correct:Store new case in the memory.


Customer Classification

• In e-c another classification problem arises: To define classes of customers with the same behavior and treat the customers according to their class:

Observations about the customer

Customer class

Customer treatment

Inexactclassification

Domainknowledge

This will be discussed in chapter 10.


The Classical CBR R4-Cycle[from: Aamodt & Plaza, 1994]

Retrieve:

Determine most similar case(s).

Reuse:

Solve the new problem re-using information and knowledge in the retrieved case(s).

Revise:

Evaluate the applicability of the proposed solution in the real-world.

Retain:

Update case base with new learned case for future problem solving.

Ret

riev

eReuse

Rev

ise

RetainCaseBase

Knowledge

NewCase

NewCaseRetrieved

Case

SolvedCase

LearnedCase

Tested/Repaired

Case

SuggestedSolution

Problem

ConfirmedSolution

This cycle shows the main activities in CBR


Retrieve: Modeling Similarity• The similarity based retrieval realizes an inexact match

which is still useful:– Useful solutions from a case base– Useful products from a product base

• Different approaches depending on case representation• Similarity measures (see chapter 6):

– Are functions to compare two cases sim: Case x Case [0..1]

– Local similarity measure: similarity on feature level– Global similarity measure: similarity on case or object level

• For special tasks (see chapters 5 and 6):– (Sub-)Graph isomorphism for graph representations– Logical inferences


Similarities (1)

• Similarities are described by measures with numerical values

• They operate on – problem descriptions, demands, products ,...

Intention: •The more similar two problem descriptions C and D are, the more useful it is two use one of the solutions alsofor the other problem. •The more similar a demand and a product are the moreuseful is the product for satisfying the demand.


Similarities (2)

• Given a fixed problem (demand) C • A similarity measure introduces a partial ordering (to be

more or less similar to C) on – the set of problems and therefore also on the case base– the set of products and therefore on the the product base.

• The basic intention means that „more similar“ also means „more useful“ with respect to the solutions

• Therefore the similarity measure controls the utility when inexact solutions are employed or the desired product is not exactly as desired available.


Similarities (3)

• An important consequence of the intention is that

similarity is strongly related to utility

• Utility is provided by the customer and should reflect

his interests and needs

• Similarity is secondary because it is used to find

solutions for the customer.

• Similarity is therefore not an absolute notion like truth

but a problem dependent notion.


Similarities (4)

• The similarity measure is the central element to navigate through the space of possible solutions or possible products.

• Instead of presenting the exact solution similarity is a concept to approximates it.

• Even when the exact or optimal solution is not available or too difficult to achieve one comes still up with at least a suggestion for the solution.


A Typical Similarity Measure

• Given two problem descriptions C1, C2• p attributes y1, ..., yp used for the representation

p

1jjj (C1,C2)simSIM(C1,C2) ω

simj : similarity for attribute yj (local measure)wj : describes the relevance of attribute j for the problem


Retrieval: Finding The Nearest Neighbor

• For a new problem C the nearest neighbor in the case base is the case (D,L) for which problem D has the greatest similarity to C.

• Its solution L is intended to be most useful and is then the best solution the case base can offer (or best available product).

• Classical databases use always total similarity (i.e. equality).

• The access to data in databases is in similarity based systems replaced by the search for the nearest neighbor. It can be regarded as an optimization process.

This requires more effort but can be much more useful.


Thresholds and Rough Sets

• The nearest neighbor (in the given case base) is not always sufficient for providing an acceptable solution.

• On the other hand, a case which is not the nearest neighbor may be sufficient enough.

• For this purpose one can introduce two thresholds and , 0 < < < 1 with the intention– If sim(newproblem, caseproblem) < then the case is not accepted;

– If sim(newproblem, caseproblem) > then the case is accepted.

• This partitions this case base (for the actual problem into three parts (so-called rough sets): accepted cases, unaccepted cases and an uncertainty set. The same works for product bases.


Retrieve: Efficiency Issues• Efficient case retrieval is essential for large case bases

and large product spaces.• Different approaches depending

– on the representation– complexity of similarity computation – size of the base

• Organization of the base: – Linear lists, only for small bases– Index structures for large bases, e.g., kd-trees,

retrieval nets, discrimination nets– 2-Phase retrieval: MAC-FAC strategies

• How to store cases or products: – Databases: for large bases or if shared with other applications– Main memory: for small bases, not shared

See chapter 7.


Reuse: How to Adapt the Solution

• No modification of the solution: just copy. Manual/interactive solution adaptation by the user.

• Automatic solution adaptation :– Transformational Analogy: transformation of the solution

• Rules or operators to adjust solution w.r.t. differences in the problems

• Knowledge required about the impact of differences

– Derivational Analogy: replay of the problem solving trace

• Complete generative problem solver

• Knowledge required about how to solve the problem in principle

– Compositional adaptation: combine several cases to a single solution

See chapter 9


Adaptation in Electronic Commerce

• The best available product may not satisfy customer’s demands sufficiently well.

• Customize the product if possible– Exchange, add or remove parts– Change parameters– etc.

• Suggest product bundles if an individual product does not satisfy the customers demands– Product Configuration

See chapter 9.


Revise: Verify and Correct Solution

• Revise phase: little attention in CBR research today – No revise phase

– Verification of the solution by computer simulation

– Verification / evaluation of the solution in the real world

– For products: Technical or customer evaluation, buying decision

• Criteria for revision – Correctness of the solution

– Quality of the solution

– Other, e.g., user preferences


Retain: Learning from Problem Solving

• What can be learned – New experience (new case)

– Improved similarity assessment, importance of features

– Organization/indexing of the case base to improve efficiency

– Knowledge for solution adaptation

– Forgetting cases, e.g., for efficiency or because out-of-date

• Methods – Storing cases in the case base

– Deleting cases from the case base

– Explanation-based learning

– Induction, e.g. of decision trees

– Neural net style learning


Advantages of CBR over other Techniques

• Reduces the knowledge acquisition effort

• Requires less maintenance effort

• Improves problem solving performance through reuse

• Makes use of existing data, e.g. in databases

• Improves over time and adapt to changes in the environment

• High user acceptance


Reduce Knowledge Acquisition Effort

CBR Systems

• Require less general knowledge

• Most knowledge in case base

• Case knowledge is easier to acquire (sometimes already available)

Solution

Problem

KBSDomain

Knowledge

Knowledge Acquisition

Knowledge Base

Traditional Knowledge-Based Systems

!!

Acquisition of general knowledge is very difficult !!


Less Effort Required for Maintenance

What is the impact of changes of the environment ?

• Rule bases or models are difficult to maintain– Many dependencies between rules– Rules of KBS often difficult to understand for non AI experts – Effects of changes of the rule base are hard to predict– Maintenance by the domain expert impossible !!

• Case bases are easier to maintain– Cases are independent from each other– Domain experts and novices understand cases quite easy– Maintenance of CBR system (partially) by adding/deleting cases– However, changes in the vocabulary container require more effort


Expert Systems vs. CBR

Expert System CBR System Expert should be replaced

Supports the inexperienceduser in routine operations

Should generate newknowledge

Searching for similar cases andadapting these if necessary

Knowledge implicitly stored inan expert system model

Knowledge explicitly stored inconcrete cases and similaritymeasures

Hard to maintain because ofunpredictable implications bymodel changes and extensions

Distinctly easier to maintainand to update

Aimed at 100% coverage ofknowledge domain

Works with the pareto principle


Databases vs. CBR

Database System CBR System simple search “all or

nothing”

using same database butsearch for most similar cases

often too many hits(underspecification) or no hitsat all (overspecification)

system can be told to showonly, e.g., 10 cases bydescending order

no specific domain knowledgeused for the search

considers domain knowledgefor search by using similaritymeasures, e.g., spatial orgeographical relations

pure database applicationscannot be used for onlineconsulting

online consulting is the powerof a CBR system


Positioning the CBR System

Closed Knowledge Model Open Database

Expert Systems CBR Systems Databases

Increasing Knowledge Centralization

IncreasingExample Orientation

Knowledge,Models

Data,Examples

Case BaseKnowledge Model

SimilarCase

NewCase

AdaptedSolution

*

*

VerifiedSolution

*

StoredSolution

RETRIEVE

REUSEREVISE

RETAIN


Search Engines / Information Retrieval vs. CBR (I)

• CBR as an alternative for traditional information retrieval systems:

“intelligent” search in unstructured documents possible

• Especially: Internet or Intranet• Someone searching will be lead to the

solution/relevant information step by step by selective questions

Search Engine

CBR


Search Engines / Information Retrieval vs. CBR (II)

PropertiesSearch Engines CBR Systems

index the Internet specific for one application

search for Web sites is possible search for problem solutions

possible search is processed by

keywords search processed by features

no specific know-how present specific know-how about the

domain used, even necessary


When to Use CBR?• ... users have to be supported and advised• ... cases can easily be identified and created,

products are simple to describe• ... easy maintenance of the case based is desired• ... no 100% coverage of the domain is required• ... Similarity based retrieval is acceptable fast

Database

ExpertSystem

InformationRetrieval

CBR

CBR technology can be understood as the fusion of these concepts whereby the advantages of knowledge-based systems are linked to existing data.


The R4-Cycle for Electronic Commerce

Retrieve:

Determine most similar product(s).

Reuse:

Adapt/configurate the product using information and knowledge

Revise:

Evaluate the product in the real world.

Refine:

Learn from customer behavior

Ret

riev

eReuse

Rev

ise

RefineProduct

Base

Knowledge

Userdemands

UserdemandsRetrieved

product

Modifiedproduct

Testedproduct

Offer

Initialdemands

Evaluation

The sales process in the CBR-view


Use of CBR Concepts in E-C (1)• In a sales context the customer has a demand to be

satisfied. The realization of the demand can be in achieved basically two ways:– 1) Use the CBR approach: Store cases of the form (demand,

product) in a case base and search for a similar demand in the base if a new demand arises.

– 2) Try to associate directly a product to a demand without storing cases.

• In the sales context several classification tasks occur (e.g. classification of customers). Again, there are the two approaches as above.

• When searching a suitable supplier also similarities can be applied


Use of CBR Concepts in E-C (2)

The classical CBR approach:

demand(customer)

product ?(class ?)

demand from case(customer from case)

product from case(class from case)

nearest neighbor search

stored

reuse



demand(customer)

product from product base(customer class)

The extended approach:

nearest neighbor search

Requires:Notion of similaritybetween• demand and product• (customer and class)



• The retrieval algorithms developed in CBR can still be applied.

• The special situations may, however, require modifications or variations of such algorithms.

• The solution transformation will play an important role because selected products often do not satisfy customer demands without modification.

• As a new requirement we will encounter configuration: There may be only parts of products available in the product base and for the demanded product one needs a design from the parts.



• Problems with the extended approach:– The terminology used in the descriptions of demand and

product (customer and customer class) is often quite different.

– Therefore similarity measures are more difficult to define.– This will be discussed in chapters 9 and 10.

• Advantages of the new approach:– There is no case base necessary (which is often not

available);– The basic techniques (e.g. similarity based retrieval, inexact

classification) can still be applied.– The similarity based approach is often superior to standard

information and data based retrieval techniques.


Advantages of Using Similarity in E-C

• Similarity based retrieval provides always an answer (e.g. an offered product).

• Answers can also be provided for incompletely stated queries.

• Change of user preferences can be reflected by changed similarity measures.

• The number of retrieved answers can be controlled by selecting the m nearest neighbors.

• Answers can be improved using adaptation.


Summary

• CBR is a technique for solving problems based on experience

• CBR problem solving involves four phases:

Retrieve, Reuse, Revise, Retain

• CBR systems store knowledge in four containers:

Vocabulary, Case Base,

Similarity Concept, Solution Adaptation

• Large variety of techniques for:

– representing the knowledge, in particular, the cases

– realizing the different phases

• CBR has several advantages over traditional KBS

• The basic techniques of CBR have been extended to the needs of E-Commerce.

chapter 3 informal introduction to similarity-based and case-based reasoning stand 20.12.00

Documents