27
P age 1 Protocols, Strategies and Architectures for Automated Negotiation Claudio Bartolini HP Labs Bristol Bologna, November 17, 2000 P age 2 Outline Part I – Definition of the Negotiation Process Part II – Importance of Negotiation in the New Economy Part III – Negotiation Protocols and Strategies Part IV – Agent-based Software Architectures for Negotiation

Outline -  · Page 3 Part I - Definition of the Negotiation Process Page 4 Negotiation A UHJXODWHGS URFHVV, aimed to the formation of an DJUHHPHQW among the SDUWLFLSDQWV

Embed Size (px)

Citation preview

Page 1

Protocols, Strategies andArchitectures for

Automated Negotiation

Claudio BartoliniHP Labs Bristol

Bologna, November 17,2000

Page 2

Outline

Part I – Definition of the NegotiationProcess

Part II – Importance of Negotiationin the New Economy

Part III – Negotiation Protocols andStrategies

Part IV – Agent-based SoftwareArchitectures for Negotiation

Page 3

Part I - Definition of the Negotiation Process

Page 4

NegotiationA UHJXODWHG�SURFHVV, aimed to theformation of an DJUHHPHQW amongthe SDUWLFLSDQWV to the process

Page 5

Negotiation Protocols

The set of rules which govern theinteraction:

• Permissible types ofparticipants (negotiators andthird parties

• Negotiation states (acceptingbids, negotiation closed)

• Events that cause states tochange (no more bidders, bidaccepted)

• Valid actions of participants inparticular states (whichmessages can be sent bywhom, to whom, at whatstage)

Page 6

Examples ofnegotiation

One to one bargaining

Auctions

Request for quotes (RFQ)

Page 7

Part II - Importance of Negotiation in the NewEconomy

Page 8

Commerce meets theWeb

First generation E-Commerce

• Amazon.com, E-Bay,Priceline.com

Second generation E-Commerce

• B2B Exchanges: e-Steel,PaperExchange

Page 9

Importance ofNegotiation

Business-to-Business tradingneeds negotiation

Negotiation can be multi-party

Difficult to automate

Page 10

Are all E-CommerceNegotiations Auctions?

The difference between auctions andnegotiation mechanisms hasblurred with the arrival of theInternet and E-Commerce

Are auctions negotiations? Arenegotiation auctions?

Mechanisms such as logrolling andsimultaneous improvements arenot easily expressed in auctiontheory

Our argument: auctions can be seenas negotiation, whereas there’smore to negotiation than what canbe expressed in an auctionframework

Page 11

Dimensions ofComplexity inNegotiation

Mechanisms

Multiple parameters of thegood/service being negotiated over

Multiple goods sold/bought

Page 12

Mechanisms

One-to-one

One-to-many (auctions, reverseauctions)

Many-to-many (double auctions)

Page 13

Multiple Parameters

Not only price

Pricing policy

Delivery procedure

Timing constraints

Payment methods

Page 14

First generation E-Commerce

Sales via the Internet haveincreased dramatically

Some companies (Amazon) sellonly via the Internet

Other companies use the net as amarketing tool

Page 15

E-bay

Most famous Internet Auction Site

Not only auctions:

• Free insurance

• Escrow service

• Traders rating service

• Authentication

• Investigation

• Dispute resolution

Page 16

Priceline.comBuyer make their own price, sellerswill match

Airline tickets, hotel rooms…

Page 17

Agents in ElectronicCommerce

First generation: Focussedprimarily on consumer trading

Second generation: realize thatnegotiation is important forbusiness to business

Page 18

Stages in Trading

What do I need?

What products provide this?

Where can I buy them?

At what price?

How do I pay?

How will it be delivered?

What post-sales support can I get?

Page 19

Shopbots and Pricebots

6KRSERWV: agents that automaticallysearch the Internet to obtaininformation about prices and otherattributes of goods and services.

3ULFHERWV -- adaptive, price-settingagents which firms may wellimplement to combat, or even takeadvantage of, the growingcommunity of shopbots.

Study by Greenwald and Kephart:what does the proliferation ofelectronic agents induce: pricewars and pricebots earning higherprofits than game-theoreticalequilibrium

Page 20

Firefly

‘Collaborative filtering’ agent

Helps decide what CD or movie tobuy

Compares your tastes with others

Proposes you try out a CD enjoyedby people with similar tastes to you

Firefly was acquired by Microsoftin 1998

Page 21

BargainFinder

The first price search agent

Finds the cheapest suppliers of agiven CD

Some traders wanted to ban it(CDLand), others welcomed it in

Page 22

SmartBidder

Simple agent in internet auctions

Bids on your behalf up to yourmaximum price

Bids just above the previous bidder

Page 23

Agent-BasedAutomated Trading

Suppliers and consumers delegateto agents

Agents negotiate with each other todetermine prices

Pricing is affected by supply anddemand

24x7 trading based on currentinformation – agent always present

Page 24

Advantages of the Web

Reduced sales overhead (noshopfront)

Potential international market

Ease of providing large amount ofinformation

Page 25

Second Generation E-Commerce

Business-to-Business (B2B) sales

General supplies (e.g. paper,electronic components)

Specialised contracts

Services (e.g. contractprogramming, translation)

Page 26

B2B E-CommerceCatalog aggregators

B2B Exchanges

Page 27

B2B CatalogAggregators

Streamlines purchasing byaggregating the productcatalogues of each supplier in oneplace and one format

Examples:

• e-Chemicals, Chemdex,Metalsite, PlasticsNet

Technology providers:

• Ariba

• CommerceOne

• Other 126 listed byB2BBusiness.net under thecategory Enablers and Builder-> Auctions and Exchanges onNovember 15, 2000!

Page 28

A Sample TechnologyProvider: Trading

Dynamics

Bought by Ariba in 2000 for 100M$

Now commercialised as AribaDynamic Trade

Fully integrated auction andexchange application

The solution allows marketparticipants to trade based on abroad range of factors, includingprice, product quality, paymentterms, service levels and deliveryoptions

Other similar solutions fromTradeAccess and Exterprise

Page 29

B2B Exchanges

Categories of B2B Exchanges

• Trading Hubs

• Post and Browse

• Auction Markets

• Fully-automated Exchanges

Page 30

Part III - Negotiation Protocols and Strategies

Page 31

Auction Theory andGame Theory

Auction theory is a type of appliedgame theory that is concerned withallocation of goods if valuation ofthe buyers for the goods areunknown

Page 32

Standard types ofauction

Ascending-bid auction (English)

Descending-bid auction (Dutch)

First-Price sealed-bid auction

Second-Price sealed-bid auction(Vickrey)

Page 33

Ascending-bid Auction

Also known as English Auction

Price is successively raised until onlyone bidder remains

That bidder wins the object at thefinal price

Its continuous version is calledJapanese Auction by someeconomists

Subject to the ZLQQHU¶V�FXUVH: theparadox that the winning bid in anauction is greater than theproduct’s market evaluation

Page 34

Descending-bid Auction

Also known as Dutch auction, socalled after the flower market

Price is successively lowered untila bidder calls out that they want theobject at that price

Page 35

First-Price sealed-bidauction

Each bidder independently submitsa single bid without seeing others’bids

The object is sold to the bidder whomakes the highest bid

The price the bidder pays is howmuch they bid for

Page 36

Second-Price sealed-bid auction

Also known as Vickrey auction

Each bidder independently submitsa single bid without seeing others’bids

The object is sold to the bidder whomakes the highest bid

However, the price the bidder paysis the price of the second highestbid

Page 37

Revenue-EquivalenceTheorem

Vickrey 1961;

Regardless of the type of auction(Dutch, English, First-pricesealed-bid or Vickrey), the highestprice paid by a group of rationalbidders is on average the same

The result is based on convenientassumptions. Under more realisticassumptions, differences in themechanisms entail difference inprices.

Page 38

Multi-unit auctions

Simultaneous auctions – shareauctions

• e.g. radio spectrum, TVfrequencies…

Sequential auctions

• Items are sold sequentially

Combinatorial auctions

• Bidder expresses preferencesfor complementary andsubstitutable items

• Complex algorithms might benecessary for winnerdetermination

Page 39

Reverse auctions

Dual case of the auction so farpresented

A single buyer, rather than a singleseller, controls the tradingmechanism

Sellers submit DVNV, rather thanbuyer submitting ELGV

Page 40

Double auctions

Buyers and sellers are treatedsymmetrically

Buyers submit bids and sellerssubmit asks

Structured process rather thanmulti-party bargaining

Example: the NYSE rule

Page 41

Agents which negotiate

An agent needs…

• A representation of thegoods/services to be traded

• An understanding of the trader’sgoals (utility function)

• A strategy for negotiation

Page 42

Utility Function

A XWLOLW\�IXQFWLRQ is a mathematicaldescription of the preferences of arational trader. The function mapsalternative choices into numericscores, such that the higher thescore, the more desirable thechoice

Multi-attribute Utility Theory analysespreferences with multipleattributes

Problems: SUHIHUHQFH�H[WUDFWLRQ�XQFHUWDLQW\�DQDO\VLV

Page 43

Negotiating Strategies

A negotiating strategy consists incarrying out the negotiationprocess so as to maximise atrader’s utility function, under theconstraints imposed bynegotiation rules and otherplayers’ behaviours

Multiple approaches:

• Rule-based

• Game theory

• Adaptive behaviour

• Genetic algorithms

Page 44

Part IV - Agent-based Software Architectures forNegotiation

Page 45

Why The AgencyParadigm Suits

Negotiation

Weak notion of Agency(Wooldridge, 1992):

• Autonomy

– No human intervention

• Social ability

– Interact with other agents

• Reactivity

– Perceive the “world” and react

• Situatedness

– Exhibit some goal-orientedbehaviour

Page 46

Issues that developersface

Wooldridge and Jennings (1995):

• Agent Theories

– What is an agent– Mathematical formalism to

represent and reason about theproperties of agents

• Agent Architectures

– From specification toimplementation of software andhardware systems

• Agent Languages

– Software communication systemsfor programming andexperimenting with agents

Page 47

Agent-OrientedSoftware Architecture

Need to define (GAIA, Wooldridge,Jennings, Kinny, 2000):

• Abstract concepts

– Roles

– Permissions– Responsibilities

– Protocols– Activities

– Liveness properties– Safety properties

• Concrete concepts:

– Agent Types

– Services– Acquaintances

• Plus: organizational abstractions(Zambonelli, Jennings, Wooldridge2000)

– Organizational Rules– Organizational Structures

– Organizational Relationships

Page 48

Example: TheFishmarket

Noriega (1997)

• Electronic Auction house

• Agent Testbed

• The Fishmarket Tournaments

Page 49

Bibliography

Negotiation

• N. Jennings, S. Parsons, C.Sierra, P. Faratin (2000):$XWRPDWHG�1HJRWLDWLRQ, PAAM2000

• G. Kersten, S. Noronha, J. Teich(2000) : $UH�$OO�(�&RPPHUFH1HJRWLDWLRQ�$XFWLRQV" –Proceedings of COOP2000 –Fourth International Conferenceon the Design of CooperativeSystems

Page 50

Bibliography

Agents in electronic commerce:

• M. Wooldridge (1992): 7KH/RJLFDO�0RGHOOLQJ�RI&RPSXWDWLRQDO�0XOWL�$JHQW6\VWHPV, PhD Thesis, Universityof Manchester

• M. Wooldridge, N. Jennings(1995): $JHQW�7KHRULHV�$UFKLWHFWXUHV�DQG�/DQJXDJHV��$6XUYH\, In Intelligent Agents(ATAL 94)

Page 51

Bibliography

Agents in electronic commerce:

• P. Noriega (1997) – $JHQW0HGLDWHG�$XFWLRQV��WKH)LVKPDUNHW�0HWDSKRU, PhDThesis, Universitat Autonoma deBarcelona

• A. Greenwald and J. Kephart(1999) – 6KRSERWV�DQG3ULFHERWV, in Proceedings ofIJCAI '99

Page 52

Bibliography

Agent-Oriented SoftwareEngineering:

• M. Wooldridge, N. Jennings, D.Kinny (2000): 7KH�*DLD0HWKRGRORJ\�IRU�$JHQW�2ULHQWHG$QDO\VLV�DQG�'HVLJQ�LQ$XWRQRPRXV�$JHQWV�DQG�0XOWL�$JHQW�6\VWHPV, Vol 3. Number 3,September 2000

• F. Zambonelli, N. Jennings, M.Wooldridge (2000) :2UJDQL]DWLRQDO�$EVWUDFWLRQV�LQWKH�$QDO\VLV�DQG�'HVLJQ�RI�0XOWL�$JHQWV�6\VWHPV, ICSE 2000

Page 53

Bibliography

Auctions:

• W. Vickrey (1962) : $XFWLRQ�DQG%LGGLQJ�*DPHV – In Recentadvances in game theory (pp.15-27) Princeton, New Jersey:The Princeton UniversityConference

• P. Klemperer (1999) : $XFWLRQ7KHRU\��D�JXLGH�WR�WKH�OLWHUDWXUH –In Journal of Economic Surveys(Vol 13-3, pp. 227-286)

Page 54

BibliographyB2B Exchanges:

• A. Sculley, W. Woods (1999)%�%�([FKDQJHV, ISI publications