2009 International Conference on Management of e-Commerce and e-Govemment

Multi-agent Systems Set for B2B E-commerce Systems

Weilin Dan, Mengqiu Song

School ofEconomies
Wuhan university of technology
Wuhan Hubei,430070, China
[email protected]

Abstract-In order to solve the condition that E-Commerce is than 100 billion in 2008. However, this mode exists the
inefficient and relies largely on labor , a B2B E-Commerce following problems:
Model (MAECS) based on the Multi-agent system (MAS) • Enterprise Information update is slow because it
was introduced. This system focused on distributed FIPA- needs professionals to publish.
Agent OS based multi-agent system, and introduced the • It is difficult for enterprise managers to spend a lot
facilitator such as buyer agent and seller agent. This system of time searching valuable information in the vast
reduced communication traffic effectively, realized the Internet, not only achieved very limited information,
asynchronous transfer of information and platform- but the accuracy and effectiveness are very low.
independent, making e-commerce system handle the business
• It is difficult for the information provider to get the
processes independently and intelligently and improved e-
commerce activities efficiency. information of the demander.
For the whole business users in the internet, this mode
Keywords-component; E-Commerce; Multi agent systems; has brought too much communication, spent a great deal of
Agent cooperative labor and material resources, and it can't get the optimal
solution. This paper presents a model of e-commerce:
Introduction MAECS. MAECS is based on multi-agent system (MAS), it
The rapid development of SCIence and technology, combines the advantages of supply chain and business
intensified competition in the market, promoted the platform as Alibaba. it can access to business information for
integration of the global economy, brought new challenges to enterprises efficiently and intelligently.
the traditional business management mode of operation. It is
the hot area of computer and management research that how I. PRELIMINARY

to make use of advanced computer technology to enhance

A. Agent
the efficiency of business management. For this reason,
many large enterprises introduced supply chain management The concept of Agent origins in 1970s, it is an offset of
(SCM). SCM is the composition of the supply chain from the distributed artificial intelligence. The most important
first and the last business as a whole, which focuses on contribution of agent technology is that it provides a new
logistics, capital flow of information sharing and operational distributed computing and solving problem idea. On the one
coordination, to achieve stable and efficient management of hand, it improves the capacity of system within the entities;
supply and demand relations. Many large international On the other hand, Agent technology provided an effective
companies such as: Hewlett-Packard (HP), Digital means to solve the problem of distributed system and
Instruments Inc. (DEC), Procter & Gamble (P & G), provide an effective conceptual model for entities'
Ericsson, etc, have been using this new method of collaboration. Agent is independent interactive, initiative and
management and thus enhanced the international response; it can not only act on itself, but also affect the
competitiveness of enterprises. However, due to technical environment. It receives feedback from the environment to
and financial reasons, the domestic small and medium re-evaluate their behaviors; at the same time, it can work
enterprises can not apply this advanced technology. At the with other Agent together. Agent System relaxed the limits
same time, Alibaba appeared as the third-party e-commerce of centralized, non-open and ordered. It provides distributed
platform for small and medium enterprises. As long as it is control, dynamic contingency processing and parallel
registered as a member of the Alibaba, it can publish processing and provides a more rapid method of problem
information online or search for the information related to solving. In addition, Agent system can reduce the cost of
their own companies and then consulate separately. Alibaba software or hardware, solving the issue of software reuse
provided effective B2B solution to the world's small and more efficiently. With all these features, Agent is more
medium-sized companies. According to statistics, as of suitable for making e-commerce systems than the traditional
2008-09-30, Alibaba market has 28.7 million registered users software method.
in China and 5.6 million registered foreign members.
B. MAS
Alibaba exports 20 billion dollars in 2008. Taobao, a C2C's
business site of Alibaba ,the transactions of which are more Individual Agent's intelligence is limited, but we can
make up agents to a system order by some organize structure,

IEEE
978-0-7695-3778-8/09 $26.00 © 2009 IEEE 3 ~computer
DOl 10.1l09/ICMeCG.2009.95 society
each Agent in order to compensate for the lack of ability to I Software I
make the whole system more than any single Agent's ability,
•
so that a multi-Agent system.
MAS refers to a number of Agent makes up a loose 1 Agent Platform

federation, those Agent service, cooperate with each other to Agent I I Agent
Management I I Directory I

• •
Facilitator
complete the same task. The members of MAS are I system

autonomous and independent, its goals and behaviors are not ~

Message Transport System
limited by other agents of the systems. They solve problems I I

and collision between their action and target by consultation I

and competition. MAS' data and resources are scattered,
Message Transport System
each member to complete the incomplete information or the I I

ability of communication. There isn't an overall control

Agent Platform
system, the task of implementation and the calculation is
asynchronous. Figure 1. The structure ofFIPA Agent platform
C. The communication and cooperation ofMAS
3) MTS (Message Transpol1 Service) provides a ACL
There are a large number of Agents in the MAS, those
agents are organized together through a loose federal, each information interaction mechanisms between different
Agent has its own wills and goals to achieve the purpose of Agents. MTS is provided by Agent Communication Channel
their activities, so the core issue of MAS is communication VICC) on platform. ACC is an entity in the platform
and coordination. providing information interaction. MTP (Message
Agent communication languages mainly are KQML Transport Protocol) is a different message interaction
(Knowledge Query and Manipulation Language) and FIPA between the ACC agreements.
(The Foundation for Intelligent Physical Agents) ACL (agent 4) AP VIgent Platform) is the physical entity can be
communication language). FIPA ACL is the main idea of deployed to run by the Agent, which includes not only the
integration of speech act theory, predicate logic and public AMS, DF, MTS, but also the underlying hardware,
ontology, to provide a standard approach to the operating system and supporting software, such as the JRE.
understanding between Agents. FIPA is a public use of the
The chart is an Agent of the FIPA communication model,
norms, it is not a specific application of technology, but for
the Agents communicate through communication protocols
different applications of GM technology; it is not a stand-
such as the Socket.
alone technology, but a set of basic skills set, allowing the
Non agent software
developers to develop the capacity of a highly complex
system synergy. The latest open standard is FIPA 2000.
FIPA ACL as a result of formal language based on SL
(Semitic Language) precise formal semantics; it is more
suitable to communication between different platforms,
heterogeneous communications than KQML.
FIPA Agent must be running in the FIPA platform, the
platform provides support for basic services, so to some
extent, can also be called Agent middleware platform, the
reference structure of FIPA Agent platform is shown in
Figure 1. SOAP CORBA

The main function of the platform is managing the life

Figure 2. The communication of MAS architecture
cycle of Agent, making agents communicate with others
agents in or outside of the platform. Agents can access to
platform services. II. SYSTEM ARCHITECTURE
FIPA management model mainly has the following
function modules: A. The federal structure
1) AMS VIgent Management System), Monitoring the There are many enterprises in MABCS, and they belong
visiting of Agent on the platform, saving AMS platform in to different sub-sectors. Therefore, The MAS of MABCS
the activities of the Agent's ID, each Agent in order to uses federal structure. The system is made up of many
obtain a valid ID must be registered to AMS at first. different federals and each federal in the system is made up
2) DF (Directory Facilitator) provides the Yellow Pages of lots of agents. Agents in different Federal can not
service on platform. Agent can register their ability or communicate directly, they communicate through a sort of
service information to the DF, it also can search from the given agent called Facilitator. The communication of agents
DFfor the information ofthe Agent providing services. transmitted to other agent by Facilitator in Federal, its
structure in Figure 3.

4
 Agent Agent
Agent aj Agent

UJUJ
Sort I Sortm

Internet

Figure 3. Federal Structure

The main task of the Facilitator is to give an abstract Seller

Buyer
description about the agent which has the service capabilities Agent
Agent
of source of information, thus playas a role of bridge
between the demand-side and the provider. The service
providers register the ability to their Facilitator, when Figure 4. The structure ofMAECS
demander inquires to the Facilitator, the Facilitator will
match the two and inform service providers the information customer relationship management (CRM); e-procurement is
related to demand-side, in order to achieve the role of a focused on the purchase of raw materials, it is similar to
bridge. In this structure, Agent communicates through the supply chain management (SCM); and the electronic trading
Facilitator, and it can not communicate directly to each other. market in the a collection of buyers and sellers of online
Without getting the features and location information of communities in order to limit the role of agent is open to
other Agents, this model can reduce communication buyers and sellers, not a single group of interests, do not own
overhead of system effectively. We assume that the number the product or service transactions, only the completion of
of users in that e-commerce systems is n and each auction the transaction platform together.
need bidding round of m, then they need take place n2 m 0
At present, the electronic trading market model includes
times auction under the model of direct communication. If three involved: the buyer (Buy-side), the seller (Sell-side)
we adopt the intermediation approach, the communication and intermediaries (intermediary) in a transaction process,
complexity is mon. so if the e-commerce system has a lot of the intermediary of the trading process provides the
numbers it must use federal model. supporting platform for communication between buyers and
In MAECS, in order to get the best possible solution we sellers, consultation, signing the contract. The relationship
use British-style auction. between the buyer and the seller is many-to-many, that is, a
buyer can get supply from multiple suppliers, and one
B. System Architecture supplier may supply a number of buyers. This many-to-many
Under normal circumstances the process of a complete manner, through competitive bidding by the general
business process should be divided into the following six optimized the buyer's purchase cost.
steps: Demand Confirmation; Product recommendations;
Sales of goods; Consultations; Purchasing and payment; e. The course ofauction
evaluation of products and services. An interaction model describes the responsibilities of an
As the main goal ofMAECS is to assure achieving inter- agent class, the services it provides, associated interactions,
enterprise supply and demand information flow, rather than and control relationships between agent classes. This
supplying chain management system requirements from the includes the syntax and semantics of messages used for inter-
funding, logistics, manpower and the overall management of agent communication and communication between agents
resources, so the functions of the e-commerce systems and other system components, such as user interfaces.
include mainly 2 parts: 1 the purchasing and payment; 2
evaluation of products and services. 1) Communication
In general, B2B e-commerce can be divided into three MAS mainly have three core algorithms in MAECS:
categories: known as the leading seller of electronic communication registration, the establishment of
distribution (e-Distribution), led by the buyer as the communication and communication.
electronic procurement (e-Procurement) and led by a third Communication registration (SCM required to register
party had known as the electronic marketplace (e-Market for the first time).
place). Electronic distribution systems is a major role in the a) Agent connected to Management Agent through.
sale of products and services, it is similar to enterprise agreement, such as the Socket.

5
 b) Agent enter password and ID. j) to query the data (query-begin?: content
c) Management Agent determine whether the Agent has (tablenameX, taskZ))
been registered, if it has been registered, refuse the
g) assessment of (ask-if ... : content (tablenameX,
registration, or to d.
d) Management Agent writes the information ofthe taskZ))
agent into the database and notifY successful registration. h) registered (Subscribe?: Content (tablenameX,
Establish communication. taskZ))
a) Agent establishes a connection with Management
Agent through agreements such as Socket.
III. AUCTION
b) Agent inputs authentication information, and
Management Agent is to find information in the database, Assume that the buyers' needs can be provided by
give error message ifthere is an error, or prompt successful several sellers (each seller can only provide one part), and
the total cost of the transaction is C, including the purchase
landing.
price of goods and the distribution cost, the basic process of
c) Management Agent put down the information and bidding as follows: a. Buyers submit Form RFP to Buyer
address ofAgent in the Agentform online. Agent through the Web to buy their goods; b. Buyer Agent
d) Management Agent will transmitte the information ask the sellers to offer prices; Buyer Agent submit it to the
issued to the Agent. Seller Agent, and add the cost of logistics, Buyer Agent in
accordance with optimization algorithm to optimize the
Communication objectives of the Seller Agent optimize the calculation of
a) Agent starts receiving thread, monitors pricing, access to an optimization program; e. If c is the
communications, and receives ifthere have. lowest, submit it to the buyer, complete the case or to b.
b) The content ofFIPA ACL message format, prompt
We can use the integer programming method of the
the error.
c) Take down the time and information ofthe above-mentioned process d to optimization algorithm.
communications into the form. Assume that buyers need to buy N different types of
d) Search the form ofonline users, send message to the commodities in the electronic trading market; and there are
users online. M sellers and L logistics providers in the electronic
2) Multi-Agent communication protocol transactions market can participate in the auction, and there
In order to facilitate the description of the inter-Agent is sufficient variety and quantity to meet the needs of the
communication protocol, we make the following agreements buyer. If the goal is to optimize the total cost ofthe buyer the
to the description ofthe problem: minimum C, then:
a) FIPA ACL content language parameters follow the N M I N M

SL. Cruin = " kP; kQ;k + L

L.JL.JX;
" LLX;ky/'I(K;kl +Kt7)Q;k
b) The communication of this article is based on the
i~l k~l l~l i~l k~l
(1)
Agent level and do not involve the physical realization.
First of all, the defmition of an ontology: (2)
N M
Term = (tableName 1, ... tableName m.task 1, ... task n)
Of which: tablename X data table ex = 1,2, ... n); task Z that IIX;Q; =Q; i = 1, 2, .... , n
the action of (Z = 1,2, ... n). i~l k~l
(3)
FIPA ACL in the written message, the form of the above
criteria are no longer used, but omitting the use of the
Of which: X;k = {a, l} .if the product i buy from the seller,
following forms: XH=l . XH=O .
°
then I otherwise I . If products 1 purchased
(performative ... ; content <expression»
Part of inter-Agent communication protocol: ykl = 1 ykl = pk . .
from the seller, then I ,else I . I IS the pnce of
a) The Executive Agent Management Agent to request k
product i which purchased form seller k. Q; is the number of
resources:(Request-begin ... : content (tablenameX, taskZ))
b) Management Agent, recognizing that resources are product i which provider by seller k. Q; is the number of
(Confirm: content (tablenameX, taskZ))
require for product i. CT; is the factor that Products for a
c) write or modifY data (Insert ... : content (tablenameX, . d 0 f time
.
. peno
certam to amend th e cost 0 f "lactor. LDTI IS .
taskZ) ((? Field] vI) ... (? Fieldn vn))) kl
. . ADT . . .
d) a proposed (propose ... : content (tablenameX, the last lime product 1 purchase. I IS the frrst t11lle that

taskZ)) from the seller to the product i purchase from seller. This is a
solution of integer programming problems, the need to solve
e) can not accept the proposed (refuse-proposal ... : k k k
content (tablenameX, taskZ)) the variable is X; ,P; ,Q; . These three variables in each
round ofthe auction process will need to solve.

6
 A buyer agent is associated with a set of customers that Of which, time for the implementation of tasks and real-
are interested in the same domain. A redder agent is time consumption; ~ for the largest amount of time, it'
associated with a site to manage the products contained in relationship with the goal is:
the site. The data structure of the buyer agent is composed of X=time 1\ 0 time=x+1 1\ D (TC W Goal 1\(-, TC ---+
three elements called Seller Catalogue (SC), Global Profile -, Goal) (7)
Set (GPS) and Profile Collector (PC). The seller catalogue In the coordination process will consume some resources,
contains, for each site E that interacted with the buyer agent because resources are limited, so to achieve the goal can not
in the past. Contains the global profiles, all the products ofE. be excessive consumption of resources for the price limits
The global profile set of all the customers associated with the for RC E TL, its relationship with the target:
buyer agent. (RC W Goal 1\(-, RC ---+ -, Goal)) (8)
As object-oriented languages, the modem enterprise can When the Agent can not be achieved that the
be seen as an organization for core business. In MAECS, performance of a specific goal of the indicator, they will
there is a task manager, whenever a new task (buy or sell) is have a desire to collaborate:
generated, there is a new task Agent, in charge of CBeli-, ((Cani(Goall\ TCI\RC)) 1\ CBeliO (IC
ascertaining the successful tender of the Agent. This
I\RC U Goal» (9)
approach may be effective in solving the bottleneck of
Let suggest there existence of the goal and objectives set
communication congestion in MAECS, to avoid or minimize
SG = (gl, g2, ... gn). If the Agent to achieve the goal of any
potential conflicts, and to achieve the maximum the interests
of the achievement of the target is equivalent to g, say gi (l :::;
of the whole supply chain. In the current study on a large
i :::; n) or sub-goals, recorded as OrSG (gi, g); if only to
number of MAS, the system communication overhead and
achieve all the goals would be to achieve the goal g, said gi
resource constraints have been organizing and forming the
(l :::; i:::; n) for the g's 'and' sub-goals, recorded as AndSG (gi,
large scale of bottleneck in MAS. Centralized coordination
g).
approach is a effective way to lower the cost of
n
communication system, this approach through the
establishment of a centralized coordination management V'lc;;-<n (OrSG(gl ,g)) B (Bel aU(achieveax;gJ
;=1
Agent for the preservation and management information of
~ (achieveayg))) (0)
all the other Agents.
n
The study on distributed coordination mechanism mainly
involves the description of the task, the task decomposition, V'lc;;-<n (AndSG(gl' g)) B (Bel an(achieveax;gJ
;=1
and the ability to assess the bids, the effectiveness of analysis
and evaluation of the implementation of the mandate.
~(achieveayg))) 00
Coordination process is divided into six stages, the TOD (Task of Domain, task-oriented domain) different
coordination model can be defmed as: from the static coordination, coordination of research in this
M=<Ag, G, P, T, S> (4) paper is a continuous interactive process, in each interaction,
Of which: Agent interaction in accordance with its strategy, it is
Ag-Coordination Agent, On the one hand, to the estimated that a variety of possible options in the future of
coordination of their needs, goals and planning; On the other interactive effectiveness and the effectiveness ofthe program,
hand is also responsible for the conditions of reception in the sum of the current, to achieve the expected effectiveness
accordance with competitors to coordinate appropriate to the principle of expected utility maximization, not only the
partners. outcome of the consultations in the sense of expected utility
G- Coordination goals. to meet the Pareto optimal, and hope to reach even close to
P- Coordinated planning. Task of completing the the global optimum.
necessary co-ordination role, the role related of the various
IV. COPYRIGHT EXPERIMENTS AND CONCLUSIONS
role and the trust relation nature between agents.
S- Coordination ofthe program, which corresponds to the Electronic marketplace model is still in the ongoing
coordination of partnerships, S E P. exploration and development, this paper starting from the
T- Collection of coordinated partnership, Ag E T. model of e-commerce technology, then give a B2B e-
Coordination objectives between Agents include two commerce system MAECS based on MAS. MAS-based e-
parts: mission objectives, that is, what kind of work should commerce software platform is one important direction of
be completed; performance targets that the Agent should be development of e-commerce software platform, as the
in a specific performance indicators to complete the work, business continues to complicate the technology of e-
such as the Agent must be given time and resources to commerce model has become increasingly high. The first
achieve mission objectives under constraints. Agent of the generation of e-commerce technology is not well adapted to
mission objectives and performance indicators can be used the current the development of activities business. Intelligent
non-modal time operator to define the timing formula. and automated e-commerce business activities is the
Tasks and objectives: inevitable trend of development, integrate the multi-Agent
GoalETL (5) technology into e-commerce software platform for the next
Performance indicators: generation of research is the development direction of the
TC=time:::;~ (6)

7
trend of e-connnerce technology will greatly promote the [3] NtgjuSDg A Concentrated-Distributed Coordination Style of
sound development of e-connnerce. Multi-Agent base on Task. (ITESS ' 2008).(943-948)
[4] Mengqiu Song. The Research of Communication in the Multi-agent
System for Supply Chain Management. Computer information, 2007,
(112-114)

REFERENCES [5] Use of an Intelligent Agent for an E-Commerce Bargaining System,

Springer-Verlag Berlin Heidelberg 2006, (300-309).
[1] Badica, Ganzha, Paprzy. Mobile agents in agents in a multi-agent e- [6] Keesoo kim, Boyd c. Developing Multi-agent E-Commerce
commerce system. Symbolic and Numeric Algorithms for Scientific Applications with JADE. Electronic Commerce Research, (200-209)
Computing. IEEE, Los Alamitos (2005).
[7] Negri, A., Poggi, A., Tomaiuolo, M., Turci, P.: Agents for e-Business
[2] Pao-Hua Chou, Kuang-Ku Chen, Menq-Jiun Wu. Bala M, Applications. In: AAMAS'06, Hakodate, Hokkaido, Japan, pp. 907-
Balachandran, Majigsuren. Trust-Based Facilitator: Handling Word- 914 (2005)
of-Mouth Trust for Agent-Based E-Commerce. Springer-Verlag
Berlin Heidelberg 2008, (293-298)