boston 16 03

24
LOGO Customer-telecommunications company’s relationship simulation model (RSM), based on non-monotonic business rules approach and formal concept analysis method. Russian Plekhanov University of Economics Victor Romanov Roman Veynberg Alina Poluektova

Upload: victor-romanov

Post on 16-Jun-2015

265 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Boston 16 03

LOGO

Customer-telecommunications company’s relationship simulation

model (RSM), based on non-monotonic business rules approach and formal

concept analysis method.

Russian Plekhanov University of Economics

Victor Romanov

Roman Veynberg

Alina Poluektova

Page 2: Boston 16 03

LOGO

Click to edit Master text stylesContents

1

2

3

4

5

6

FCA for rules mining

Business rules application at telecommunication sector

EDM-conception and business rules application technology for decision making

Business rules theory

BRMS review

The problem actuality

Page 3: Boston 16 03

LOGO

Click to edit Master text styles

Why business rules?

Dynamic competition economy

In big and medium business a lot of documents contain business rules.

It is difficult to find and change them

EDM new conception propose extract business rule as

different component,

This makes possible more easy update

them

Page 4: Boston 16 03

LOGO

Click to edit Master text styles

What business rule is

static void processLoanRequest(Borrower borrower, Loan loan) { System.out.println("Processing request from " + borrower.getName()); // Approve or reject the loan checkLoanConditions(borrower, loan); // Display the verdict if (loan.isApproved()) { System.out.println("==> Loan is approved :-)"); } else { System.out.println("==> Loan is rejected :-("); for (Object msg : loan.getMessages()) { System.out.println("==> Because " + msg); } } } /** * Check conditions on the borrower and the loan using hard-coded policies */ static void checkLoanConditions(Borrower borrower, Loan loan) { // Check maximum amount if (loan.getAmount() > 1000000) { loan.addToMessages("The loan cannot exceed 1,000,000"); loan.reject(); } // Check repayment and score if (borrower.getYearlyIncome() > 0){ int val = loan.getYearlyRepayment() * 100 / borrower.getYearlyIncome(); if ((val>=0) && (val<30) && (borrower.getCreditScore()>=0) && (borrower.getCreditScore()<200)) { loan.addToMessages("debt-to-income too high compared to credit score"); loan.reject(); } if ((val>=30) && (val<45) && (borrower.getCreditScore()>=0) && (borrower.getCreditScore()<400)) { loan.addToMessages("debt-to-income too high compared to credit score"); loan.reject(); } if ((val>=45) && (val<50) && (borrower.getCreditScore()>=0) && (borrower.getCreditScore()<600)) { loan.addToMessages("debt-to-income too high compared to credit score"); loan.reject(); } if ((val>=50) && (borrower.getCreditScore()>=0) && (loan.getAmount() >

Applications codesIT

Business Logic

Business

Business rule is the assertion at the natural or formallanguage, which for each state of business system defines permissible decisions on business control

Page 5: Boston 16 03

LOGO

Click to edit Master text styles

The main BRMS vendors :

IBM ILog Jrules

FICO Blaze Advisor

Corticon BRMS

Innovations Software Technology Visual Rules

The Forrester Wave™ за второй квартал 2008 г.

Page 6: Boston 16 03

LOGO

Click to edit Master text styles

Business Rule Management System

The sources where rules originated from

Applications

Processes Personell

Documents

Business rules management system

The rules are stored and updated

The rules are extracted and executed

The rules are inserted

User Applications

Rules repository

Rules Server

Rules+

Metadata

Page 7: Boston 16 03

LOGO

Click to edit Master text styles

BRMS Applications Domains

Telecommunication sector

Banking

Management

State management

Insurance

Medical Care

Page 8: Boston 16 03

LOGO

Click to edit Master text styles

Corticon Technologies: Corticon BRMS

Software Environment

Data Base Connector

Rules ModelingServer

Rules Execution Server

Page 9: Boston 16 03

LOGO

Click to edit Master text styles

IBM’s ILog JRules

Architecture ILog JRules

Business rule development

Rule project

Object Model

Rule parameters

Vocabulary

business rule, decision tables

Flow rule

Business Rule management

Rule repository

Web application

Synchronize Synchronize

Deploy

DeployDeploy

Applicationrepository

Applicationrepository

Applicationrepository

Decision Validation Services

Decision Validation Services

Page 10: Boston 16 03

LOGO

Click to edit Master text stylesRule Studio Rule Team Server

Rule Execution Server Rule Solutions for Office

Component of ILog JRules

Page 11: Boston 16 03

LOGO

Click to edit Master text styles

Innovation Technologies: Visual Rules

Analysis Modeling

Monitoring

Execution

Deployment Administration

Documentation

Test and Simulation

Page 12: Boston 16 03

LOGO

Click to edit Master text styles

FICO: Blaze Advisor

Architecture of Blaze Advisor

Testing Rule Repository

Rule Development Repository

Business Rule Authoring

Rule Development

Custom

ers A

pplica

tion

Deploym

ent Manag

er

Application Server

Production Rule Repository

Page 13: Boston 16 03

LOGO

Click to edit Master text styles

Business rules application for business system decision making1

1 Business rule based data analysis for decision support and automation http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.94.928&rep=rep1&type=pdf

Page 14: Boston 16 03

LOGO

Click to edit Master text styles

The business rules formal definitionAt the theory level of first level logic (FOL) business rules have

statement view IF-THEN and expresses logical consequence or implication.

IF (conditions), then (the list of actions),

else(alternate list of actons).

p is a assertion, named as

antecedent, which is describing state of business conditions

q – assertion named as consequent, describing

decision which are offering in this conditions.

IF p, THEN q, where

Page 15: Boston 16 03

LOGO

Click to edit Master text styles

Business rules development and management

Page 16: Boston 16 03

LOGO

Click to edit Master text styles

Formal Concept Analysis Formal context K:=(G,M,I) consists of sets G,M and a

binary relation I G ×M⊆ . M –attribute set, G –objects sets (g,m) I ∈ - object g has attribute m Let us define the mapping:

ϕ: → и ψ: →

ϕ(A)=def {m M | gIm g A}, ∈ ∀ ∈ ψ(B)= def {g G | gIm m ∈ ∀ B}, A G, B M.∈ ⊆ ⊆

If A G, B M⊆ ⊆ , then

(A,B)- formal concept of context K, if ϕ(A) = B, ψ(B) = A

G2 M2 M2 G2

Page 17: Boston 16 03

LOGO

Click to edit Master text styles

Formal Concept Analysis (FCA)

Subconcept - superconcept relationship:

A1,A2 G, B1,B2 M:⊆ ⊆1. (A1,B1) (A2,B2) (A1 A2)⊆ (B2 B1)⊆2. (A1,B1) – subconcept,3. (A2,B2) – superconcept, A1,A2 – intentB1,B2 - extentThe relationships above define concepts lattice

Page 18: Boston 16 03

LOGO

Click to edit Master text styles

Formal Concept Analysis

FCA may be used for visualization telecommunication company’s customer groups, that make possible for management assign to these groups corresponding set of discount options.

Besides selecting the group of customers FCA method provide possibility by mean data mining approach extract new rules from customer database.

The clients may be considered as an objects and their personal data, realty employment positions may be regarded as attributes.

According to these data subsets of groups and their attributes may be selected as a concepts with common features.

Page 19: Boston 16 03

LOGO

Click to edit Master text styles

CUSTOMER’S ATRIBUTESCUSTOMER’S ATRIBUTES

PERSONAL DATA sitizen agegendersingle

JOB DATA studentheadInc_hInc_mInc_l

SERVICE CONSUMPTION DATA sms \mmsLoc_callInt_callConfGprs

ACTION EFFECT Act1_eff Act2_eff Act3_eff

Page 20: Boston 16 03

LOGO

Click to edit Master text styles

Formal context “customers” Context part 1

Context part 2 (continuation)

Page 21: Boston 16 03

LOGO

Click to edit Master text styles

Concept lattice “customers”

Page 22: Boston 16 03

LOGO

Click to edit Master text styles

The rules discovered The rules discovered by FCA look like this: different kind

of if customers satisfy different conditions and for them different marketing actions are effective:

< 1 > age_25 gender_male head sms \mms Loc_call Gprs Cons_mid ==> Inc_m Act1_eff;

IF age <= 25 AND gender_male = true AND head = true AND sms\mms = true AND Loc_call = true AND Gprs= true AND Cons_mid= true THEN Act1_eff;

< 2 > age_25 single Loc_call Cons_mid ==> Act2_eff;

IF age<=25 AND single = true AND Loc_call = true AND Cons_mid = true THEN Act2_eff;

< 3 > Cons_high ==> sitizen Inc_h Int_call Gprs Act3_eff;

IF Cons_high = true AND sitizen = true AND Inc_h = true AND Int_call = true AND Gprs THEN Act3_eff;

Page 23: Boston 16 03

LOGO

Click to edit Master text styles

The rules with confidence <100 %

63 < 5 > Cons_mid =[80%]=> < 4 > Act2_eff;

66 < 5 > single Loc_call =[80%]=> < 4 > Act2_eff;

Page 24: Boston 16 03

LOGO

Click to edit Master text styles

Rules quality criteria Let M – attribute set and G objects set. The rules are defined as the implication X⇒Y,

where X,Y M, X ⊆ Y =. The implication means that all objects of context which contain attributes X also

contain attribute Y. That is in the situation X manager ought make decision Y.

support

Supp=card(ψ(X)/ card(G)) - is a rate of context

objectsK := (G,M, I),

which contain attributes X

Is defined as conf( X⇒Y)=supp(XY)/

supp(X)

3

Is defined as supp(X Y)/

supp(Y) supp(X)

confidence lift conviction

Conviction conv(X⇒Y)=1-

supp(Y)/1- conf(X⇒Y)