On the Use of Optimization Techniques for Strategy Definition in Multi Issue Negotiations

ON THE USE OF OPTIMIZATION TECHNIQUES FOR STRATEGY DEFINITION IN MULTI ISSUE

NEGOTIATIONS

Κυριακή ΠαναγίδηΚυριακή Παναγίδη

Επιβλέπων Καθηγητής: Ευστάθιος Χατζηευθυμιάδης

Εθνικό και Καποδιστριακό Εθνικό και Καποδιστριακό Πανεπιστήμιο ΑθηνώνΠανεπιστήμιο Αθηνών

1/

Contents

Definitions Electronic Commerce Intelligent Software Agents Electronic Marketplaces Negotiations

Problem DefinitionStrategy DefinitionProposed AlgorithmsExperimentsConclusions Future Work

1/

“Optimizationas ageless

as time” …

Electronic Commerce

Electronic Commerce (E-Commerce) is defined by the Electronic Commerce Association as: “any form of business or administrative transaction or information exchange

that is executed using any information and communications technology” . “business practice related to buying and selling goods, products or services, in

the Internet”

Consumer Business

Consumer Consumer-to-Consumer

Example: Ebay

Consumer-to-Business

Example: PriceLine

Business Business-to-Consumer

Example: Amazon, Dell

Business-to-Business

Example: IBM, SAP

1/1/

Intelligent Software Agents - IAs

Intelligent software agents are programs acting on behalf of their human users”

“Intelligent software contains features as perception, interpretation of natural language, learning and decision making”

“A piece of software which performs a given task using information gleaned from its environment to act in a suitable manner so as to complete the task successfully. The software should be able to adapt itself based on changes occurring in its environment, so that a change in circumstances will still yield the intended result.”

“Software agents carry out certain operations on behalf of a user or another program with some degree of independence or autonomy combined with a set of goals or tasks for which they are designed”

“Intelligent Agents are computerized servants, it is software that communicates, cooperates and negotiates with each other. They have the ability to take over human tasks and interact with people in human like ways. They are bringing technology into a new dimension simplifying the use of computers, allowing humans to move away from complex programming languages creating a more human interaction” As a third

partyDegree of reasoning

1/

Intelligent Software Agents - IAs

Characteristics what the user needs

how is going to satisfy user

the IA should have the ability to modify the human user requests and ask for additional information or clarifications

Accept the user’s statement of goals and carry out the task delegated to it

take initiatives

Try to do what is asked for and act in order to achieve the user’s goals

recognize the user’s preference

interact with other IAs, programs or

humans

dynamically assess which actions to

execute and when

1/

Intelligent Software Agents- IAs

Types

CollaborativeIAs

Mobile IAs

Personal IAs

Network IAs

Desktop IAs

ApplicationDomains

Adaptive UserInterfaces

E-Commerce

Workflow andAdministrativeManagement

Information Accessand Management

Mail andMessaging

Collaboration

Mobile Access/Management

Systems andNetwork

Management

1/

Intelligent Software Agents- IAs

Barriers:• IAs should have access to their catalogues. • User goals have to be specified. • Users have to obtain information such as prices,

product’s issues, returning policies, delivery time, • Security problems may occur when submitting

sensitive information

1/

Electronic Marketplaces

“Virtual location where entities that are not known in advance can cooperate in order to achieve common goals. These entities have their own preferences and strategies”

Most of the proposed E-marketplace’s models are classified in the following two categories:

1. Direct transactions among providers and consumers2. IA-based brokered transactions

1/

Automated Negotiations

“a decentralized decision-making process used to search and arrive at an agreement that satisfies the requirements of two or more parties in the presence of limited common knowledge and conflicting preferences.”

“the process where entities try to agree upon the exchange of a product or as a mean of compromise, in order to reach mutual agreements.”

1. Electronic automated negotiation systems (EANSs)2. Negotiation support systems (NSSs)

1/

Electronic Negotiations

Properties in Automated Negotiations:1.Simplicity2.Efficiency3.Distribution4.Symmetry5.Stability6.Flexibility

CheckMarket

situation

Decidewhat to do

Search foroffers

Search foroffers

Post an offerand wait untila counter offer

Make acounter-offer

Make acounter-offer

Negotiate

Want tocomplete thenegotiation?

Complete the negotiation

No action

Start

Buy Sell

No offers

No

Yes

1/

Electronic Negotiations

1/

ManipulationStrategies

OptimizedPatient

Patient

Desperate

UsersStrategy

Argumentation

current bestsolution

opponent’sbehavior

Decisionselection

Mechanismfollowed

Number ofparticipants

Issuesinvolved

Negotiations

Many-to-Many

One-to-Many

One-to-One

Clearing-Middle member

Driven

Auctioning-Seller Driven

Bidding-Buyer Driven

Bargaining-Buyer driven

Many

Single

Electronic Negotiations-Problems

Real Life Negotiation ProblemsReal Life Negotiation Problems

Ill definedInformation not

equally distributedParticipants with

partial knowledge Communication is

ambiguous or imprecise

Complexity of Human behaviorComplexity of Human behavior

Multiple issues negotiation

Similar product suggestion

Correlated product suggestion

UltimatumNegotiation costLearning

1/

Problem Definition

Simultaneous

No Coordinator

No knowledge

Buyer Driven

One-to-many

Problem Definition

Product has a number of issues that increase or decrease each

player’s utility. An example : Price Delivery time Quality of Service (QoS) Seller’s trust

Problem Definition

Simultaneous

No Coordinator

No knowledge

Multi Issue

Buyer Driven

One-to-many

Problem Definition

Goal : Choose the best agreement

Problem is rising:“How do we evaluate two or more deals with

different issues/sets?”

1/

Problem Definition

Buyer i is in “worst case”1. Price2. QoS

1/

Problem Definition

“How do we evaluate two or more deals with different issues/sets?”

Answer: Utility

Restrictions:1. Proportional/ Not Proportional2. Ultimatum

1/

issuesmj

buyerssubNi

jv

m

jj

wiU

0

0 ,

1

Weights DefinitionWeights Definition Space ConvergingSpace Converging

Solve our problem like a mathematical problem, in which we change the weights of issues involved in negotiation

Studied algorithms:HeuristicSimplexAnalytical Hierarchy Process

Strategy Definition

1/

Solve Like in nature we assume our buyers like particles moving in space

Studied algorithms: Combination of Particle Swarm Optimization and Virtual Force

Weights Definition- Heuristic Method

Comparison between the values of issues of buyeri and the values of issues of the agreement.

Each issue then is characterized as an issue that needs a change or not

CI cI

ciI ciI

Weights Definition- Simplex Method

Maximize

InputOutput

Vi, Vagreement Wi

n)1,2,...,(j

0j w3.

n

1j agreementUjwjo .2

n

1j1jw 1.

j

wn

1j jo

Restrictions

Weights Definition- Analytic Hierarchy Process

A=

Feature’s Name Min Max Negotiabl

e

Proportionate Value

Price 10 100 True False 60

Trust 0 1 False True 0.6

Delivery 0 10 True False 5

Relevancy 0 1 True True 0.6

Weights Definition

Space Converging-PSO with VFA

o Uses a number of IAs

(particles) that constitute a

swarm moving around in the

search space looking for the

best solution

o Each particle in search space

adjusts its “flying” according

to its own flying experience as

well as the flying experience

of other particles

Space Converging-PSO with VFA

Each particle adjusts its travelling speed dynamically corresponding to the flying experiences of itself and its colleagues

Each particle modifies its position according to:

• its current position

• its current velocity

• the distance between its current position and pbest

• the distance between its current position and gbest

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Space Converging-PSO with VFA

Particles = Buyers bargaining a set of productCannot be presented by a set o 2 coordinates (x,y)VFA algorithm

Every product is a vector [V1,V2,…Vn]Particle is moving in N-dimensional space

Space Converging-PSO with VFA

next position xi(t+1) depends from the velocity vi(t), which is equal to

where and c1, c2 are random generated values.

Price

QoS

Lb

Global best (Gb)Local best (Lb)

Current Position(CP)

CP

Gb

NP

Next Position (NP)

(ti

xgi

(P2

c(t))i

xli

(P1

c(t)i

v

Experiments - Performance Metrics

The agreement ratio (AG)

Average Buyer Utility (ABU)

Average Seller utility (ASU)

R

SNAG

||

)max( iF UU ||

||

1

SN

UABU

SN

kFk

||

||

1

SN

UASU

SN

kSk

Experiments - Performance Metrics

Average Rounds (AR)

Number of successful thread (Pt)

Fairness (F)

),min(

*

sb TT

tAR

H

HSH

||R

SHP

R

kk

t

1

cV

cVp

F

|2

|2

*

Experiments

Set of experiments1. 300 negotiations NT = 50, I = 4 and V in [10, 300]

(450.000)2. 300 negotiations , V = 100 NT = 50 and I=2k, where

k=2,…,5. 3. 500 negotiations: V = 100, I = 4 and NT 5 in [5,50].

*Seller’s cost is randomly selected in the interval [10, 50].

Experiments- AG

Experiments- ABU

Experiments- ASU

Experiments- AR

Experiments- Pt

Experiments- F

Conclusions

The basic idea :an algorithm which can deal with one-to-many, concurrent, dynamic with limited knowledge negotiations

Heuristic, Simplex and AHP methods, redefine the weights of product’s

Moving IAs in the N-dimensional space applying the Particle Swarm Optimization algorithm (PSO) combined with VFA.

The average utility gained by the buyer in all methods is above 50%.

PSO algorithm can handle excellent a large number of issues and a large number of IAs.

Future Work

Relevant function for dynamically change of weights for the seller’s part

The following step for PSO algorithm is to study whether the behavior of particles will change, if the weights of issues can be dynamically defined again during the negotiations.

The comparison of our results with real data would give us more realistic perspective between the developed methods providing us with the “closest-to-human-behavior” methodology

Thank you for your attention!

27/281/

Questions;

28/28