A Shopping Model in Agent-mediated Electronic Commerce

A Shopping Model in Agent-mediated Electronic Commerce
Tan
Xueqing , Zeng
Ziming
School of Computer Science, School of Information Management,Wuhan University,P.R.China ,430074
Abstract The prosperity of electronic commerce has changed the traditional trading behaviors and
more and more people are willing to conduct Internet shopping. This paper proposes a personalized
shopping model, which makes use of agent technology to enhance the automation and efficiency of
shopping process in Internet commerce. Based on the agent technology, the shopping model integrates
knowledge-based decision-making method and preference acquiring method to provide decision support
for automatic shopping. In order to maintain a conversation with multi-suppliers, the commodity
ontology is also utilized to support sharable information format and representation. Finally, an
experimental prototype system for used car shopping is developed for demonstrating the proposed
shopping model. The conclusions show that the system performs efficiently and can help customers save
enormous time for Internet shopping.
Keywords
Agent; Electronic commerce; Knowledge-based decision; Preference acquiring;
Commodity ontology
1 Introduction
In the recent years, the rapid advances of Internet and Web technologies have promoted the
development of electronic commerce. Through the Internet, different customers and suppliers can now
easily interact with each other and have their transactions within a minimum time. Forrester research,
International Data Corp., and Nielsen Media Research have reported that the number of people buying,
selling, and performing transactions on the Web are increasing at a phenomenal pace [1]. However, the
exponentially increasing information along with the rapid expansion of the business Web sites causes the
problem of information overload. On the other hand, the potential of the Internet for transforming
commerce is largely unrealized. E-purchases are still largely non-automated. There are no proper
mechanisms to facilitate electronic transaction and automate shopping process on behalf of customers.
So a human buyer is still responsible for gathering commodity information from multiple suppliers on
Internet, making decisions about each commodity, then making the best possible selection, and
ultimately performing the e-payment.
Software agent technologies [2] provide a new scenario that is used to develop the new-generation
e-commerce system, in which the most time-consuming stages of the consumer’s shopping process will
be automated. According to Guttman [1], agent-mediated commerce systems based on consumer buying
behavior mode should has six stages: (1) need identification, (2) product broking, (3) merchant broking,
(4) negotiation, (5) purchase and delivery and (6) product service and evaluation. Currently, Some
agent-mediated prototype systems have been developed to automate some of these stages. Firefly [3] is
the first generation of shopping system, which is mainly designed for price comparison shopping and
pops up a list of commodities with price as a result to customers. This system only supports product
brokering stage. Kasbah [4] is an agent-based market system developed by MIT Media lab, in which
customers my assign the task of buying or selling a specified good to an agent which then performs
negotiation and settlement of deals according to the user’s choice. This system supports merchant
brokering stage and negotiation stage. Tete-a-Tete [5] is an agent-mediated comparison shopping system
that allows customers to negotiate across multiple terms with merchants. It uses integrative negotiation
interaction model and supports the stages of product brokering, merchant brokering and negotiation.
These agent-based systems, however, cannot support: (1) need identification, and (6) product service and
evaluation of consumer buying behavior.
In this paper, we propose an agent-based shopping model for electronic commerce. The aim of it is to
automate shopping process by assisting customers to have commodity information retrieval and
comparison in the massive information environment of the Internet. The paper is organized as follows. In
section 2, we analyze the relevant problems about the intelligent shopping system and the shopping
849
process is described. In section 3, we implement the agent-based shopping model. Finally, the
conclusions are drawn in section 4.
2 The Analysis of Intelligent Shopping System
2.1 Relevant Problems
The aim of the system is to automate shopping process by assisting customers to have commodity
information retrieval and comparison in the Internet. To automate the customer’s shopping process, the
system should solve the following problems:
(1) How to build the comparison mechanism: In order to reduce or eliminate customer intervention
during shopping activity, the system should have autonomous transaction facilities. The system generates
the commodity shopping lists then gathers commodity information from multiple suppliers on Internet.
The system should have comparison mechanism that helps the customers make the best possible
selection of supplied commodities.
(2) Adaptation to customer preferences: The system should reason about the customer’s personal
preferences automatically from his/her shopping history and evaluation of commodities that he/she has
purchased.
(3) How to implement multiple supplier access: the shopping system should gather commodities
information from multiple suppliers automatically. In order to maintain a conversation between the
shopping system and suppliers, there should be a common language to support shared data format and
representation about the commodities information.
2.2 Solution
In order to solve the problem above, we have designed the shopping system with agent technology.
Based on the agent technology, the shopping system integrates knowledge-based decision-making
method and preference acquiring method to provide decision support for automatic shopping. The
shopping process is list as followers and its workflow is shown in figure 1.
(1) First, the shopping system performs multiple suppliers searching task. The commodities can be
collected from the suppliers by a search engine and stored in the internal commodity database.
(2) After the system gets all the commodities information, it asks the customer answer some
qualitative questions to collect his/her needs about the commodities.
(3) After gathering the consumer’s qualitative needs, the system can obtain the built-in expert
knowledge to calculate the optimality of each commodity using multi-attribute decision approach.
(4) Once the currently available commodities have been ranked, the commodity with the top rank will
be recommended to the customer as the candidate.
(5) In addition to the above expertise-based approach, the system will reason about the customer’s
personal preferences from his/her purchasing history. Based on the customer’s up-to-date shopping
preferences, the system can provide more candidates to the customer.
(6) After the customer has the decision to buy, the shopping system stands for him/her to perform
e-payment and the process of shopping is completed.
3 The implementation of the Shopping Model Based on Multi-agent
3.1 Intelligent Agent and JADE
Agent technique is one of the most important technologies developed to support the Internet
applications. According to Wooldridge, agent is the software entity that is situated in some environment,
and is capable of autonomous action in this environment in order to meet its design objectives. The basic
properties of agents are following: (1) Reactive, (2) Pro-active, (3) Autonomous, (4) Object-oriented, (5)
Social ability. One alone agent can be useful and execute the particular tasks on behalf of its user.
However, in the most of cases one agent exists in environments that contain other agents. So multi-agent
systems (MAS) can be formed, in which agents can interact with each other through the message
passing. These agents are presented some kind of mechanism to collaborate with each other.
850
begin
Searching the commodities
from the suppliers
Collecting the custom er ‘s
qualitative needs about the
comm odities
Collaborative
approach
Reasoning the custom er’s
personal preferences
Calculate the optimality of
each comm odity with the
built-in expert knowledge
Providing the second
candidate to the custom er
Selecting the comm odity
with top rank as candidate
The customer has the
decision to what to buy
Performing e-payment
Shopping is completed
Fig 1. Shopping process of the system
When a group of individual agents forms a MAS, two issues related to the design of MAS should be
addressed if agent technology is to be widely used. One is agent communication language that allows
agents to interact each other while they hide their internal work details by exchanging information and
knowledge. Examples of those ones are KQML(Knowledge Query and Manipulation Query and
Manipulation Language) and FIPA-ACL (Foundations for Intelligent Physical Agents-Agents
Communication Language). The other is agent development platform. There are various agent platforms,
such as IBM Aglets, ObjectSpace Voyager, the Agent Builder Environment and the General Magic
Odyssey Agent System etc. These platforms provide an effective framework for the dispatching,
communication, and management of multiple agents in the Internet environment.
For our research work, however, we have employed JADE (Java Agent Development Framework) [6]
as the prototype development tool. JADE is in compliance with the FIPA specifications. It supports most
of the infrastructure related to FIPA specifications, such as communication language protocols, message
encoding, and white/yellow pages service.
3.2 The Architecture of Multi-agent based System
As analyzed in section 1, the shopping system in our work is to create an interactive environment in
which the customer can express his need to the system, and the shopping system can use the ephemeral
information from the customer together with the built-in expert knowledge to find suitable commodity
for him. On the other hand, in order to give the customer more choice space, the system should learn the
customer’s past shopping preference automatically and make the second recommended commodity for
the customer. Therefore, the overall goal of the shopping system includes two aspects: (1) analyzing
the customer’s requirements and finding out the most suitable commodity with the optimal quality; (2)
learning to the customer’s personal shopping preferences automatically and providing the second
candidate for the customer.
To achieve the above goal, the system we developed for used car shopping mainly include five agents:
an interface agent for interacting with the customer, a buyer agent for searching the commodities from
851
the suppliers, an expert agent for embedding external expert knowledge for internal use, an evaluation
agent for calculating the optimality of each commodity, and preference agent for learning customer’s
preferences. The overall system architecture is shown in Figure 2.
customer
user
database
Preference
Agent
Market Place
Interface Agent
expert
knowledge
Evaluation Agent
Expert Agent
commodity
database
Buyer Agent
Fig 2. The architecture of the shopping model
(1) Interface agent
The main work of the interface agent is bidirectional communication between the shopping system
and customers. In order to collect and analyze the customer’s personal needs, the interface agent asks
him some specially designed questions about the commodities. In the shopping model, assuming that the
customer does not have enough domain knowledge to answer quantitative questions regarding the
technical details about the commodity, the system has to inquire some qualitative ones instead. For
example, the system will ask the customer to express his need on the mechanical feature rather than the
type of engine and steering column. In our developed prototype system, the customer’s qualitative needs
include: mechanical features, braking effect, safety consideration, lighting effect, convenience for
steering, comfort effect. After collecting the consumer’s qualitative needs, the interface agent can then
deliver them to the evaluation agent, which calculates the optimality of each commodity.
(2) Buyer agent
Buyer agent is a mobile agent, which can migrate to the marketplace and search for the commodity
information from multiple suppliers. When it finds one supplier, it will ask for offers about the
commodity from the respective supplier. The offers information has the following format (Figure 3),
where ‘Cat” represents the category of commodity that a customer want to buy, ”Prod-i “ is a
commodity offered by the ith supplier and “Sup-i” the ith supplier. After the buyer agent get all offers, it
will return back and store the commodity information in the internal commodity database.
C at
P rod A
S up A
P rod B
S up B
...
F ig. 3. M essa ge form at for offer req u es t
(3) Expert agent
As is indicated, an important issue in the design of the system is how to use the expertise to provide
the knowledge-based decision support. The expert agent provides the communication interface with
human experts, by which the experts can embed their personal knowledge into the system and give a
score of a commodity in each qualitative need defined before. With the expert agent, the system can
collects opinions from different experts to give more objective suggestions. Then the expert agent will
convert them into a specially designed internal form for knowledge representation. However, human
experts seldom reach exactly the same conclusions. They may give different scores of the same
commodity in the same qualitative need since their preferences are different. In order to resolve this
problem, the system synthesizes all the expert’s opinions and assigns the same weights for them in the
system implementation. In this way, the expert agent can transfer each commodity to a rank form and
852
calculate its optimality accordingly.
(4) Evaluation agent
The evaluation agent is an important component of the shopping system. It uses a multi-attribute
decision making approach, which is derived from TOPSIS [7] (technique for order preference by
similarity to ideal solution), to calculate the optimality of each commodity for the customers. A
commodity C i can be expressed as a vector of C i =< f1 , f 2 ,.., f n > , each value f j represents the
relative performance of the commodity in the qualitative feature i . In this way, the optimality of a
commodity can be measured by the equation:
F=
F−
F+ +F−
(1)
n
where
F+ =
∑[ω
( f j − f j _ best )]2
(2)
( f j − f j _ worst )]2
(3)
j
j =1
n
and
F− =
∑ [ω
j
j =1
In the above equations, n is the number of commodity qualitative features; f j is the normalized
performance value of a commodity in the feature dimension j ; f j _ best and f j _ worst are the best and
worst performance value (normalized) in the same dimension, respectively; and
ωj
means the
customer’s relative need in this feature. The above measure is based on the principle that the selected
solution should have the shortest distance to the ideal solution. Based on the above measurement, the
evaluation agent will calculate the optimality of all the commodities and select the commodity with top
score as the candidate, then recommend it to the customer via interface agent.
(5) Preference agent
In order to give the customer more choice space and speed up the shopping process, a collaborative
approach is used in our work to make the second recommended commodity for the customer and will be
implemented by the preference agent. The preference agent will reason the customer’s purchasing
preference from the User Database (UserDB) and perform preference analyzing. UserDB records the
customer user profile and customer transaction records. The preference agent will trace the customer’s
recent shopping behaviors on Web, tracking the customer’s shopping history so as to learn to know the
customer’s preference. This approach is intelligent involved but requires Web technology, data mining
techniques, and sophisticated mapping functions to map acquired information to customer’s preferences.
According to the customer’s up-to-date preference, the preference agent can learn to know the
customer’s actual needs in different qualitative feature dimensions, and the acquired preference is used
to find new candidate again.
3.3 Commodity Ontology
The shopping system should gather commodities information from multiple suppliers, however, it is
difficult to exchange information between the shopping system and the suppliers because of the different
commodity data format in database and representation. In order to maintain a conversation between the
shopping system and suppliers, there should be a common language to support shared data format and
representation about the commodities information. This is established by means of an ontology, which
contains the main concepts owning to the domain we are dealing with. In addition to this information,
the ontology also includes attributes, values, relations between concepts and axioms so that consistency
checking and inferences are done. Therefore, the main ontological entity in the prototype system
developed in our work is the concept, but the use of other ontological entities such as attributes is also
possible in the model in order to provide the system with powerful representation capabilities.
A simple example of commodity ontology relating to the user car domain can be described as
follower.
853
Concept:
< concept comment= “ “ name = “CAR”>
< alternative-names>
< name > AUTOMOBILE < /name >
< name >AUTO < /name >
< name > MOTOCAR < /name>
< /alternative-name>
< specific-attributes >
< attribute comment = “ “ name = “MAKER” type=”STRING”>
< /attribute>
< attribute comment = “ “ name = “ MODEL“ type = “STRING” >
</attribute >
< attribute comment = “ “ name = “STANDARD FEATURE DESCRIPTION” type
=”STRING”>
< attribute comment = “ “ name = “ GUARANTY“ type = “STRING” >
</attribute >
< attribute comment = “ “ name = “ PRICE“ type = “STRING” >
</attribute >
</specific-attributes>
</concept>
Relation:
<relation name=”COMPONENT_OBJECT”>
<concept_name> ENGINE
</concept_name>
<concept_name> TRANSIMISSION DEVICE
</concept_name>
<relation_type>
<property> NonSymmetry </property>
</relation_type>
</relation>
It can be noticed that every concept, besides the “main name”, has also alternative names in addition
to its attributes. In this case, in addition to the concepts taking part in the semantic relation under
question, the relation will have a name with the relation type and eventually some other properties
associated to that relation.
3.4 Web Application
With the purpose of applying intelligent agents in a real environment, JADE platform should be
integrated into the Web application. The application framework is shown in figure 4, which consists of
two important components.
One component is agent integration, which employs JADE platform together with the above-referred
agents in section 2.2. At first, the environment initialization is needed in order to start working with
JADE. This process can be implemented by AgentLoader, which reads configuration files and creates
the AMS (Agent Management System) and DF (Directory Facilitator) agents. Then AMS and DF agents
provide white/yellow pages services respectively. On the one hand, DF provides a yellow pages service
to the other agents in the system, which executes the tasks of agent registration and lookup. When each
agent is created, it should be registered in AMS. On the other hand, DF is responsible for monitoring the
life cycle of each agent and tracing the behavior of it. In this way, all the agents in the system can be
effectively managed and their inter-communication will be facilitated well.
Another component is application design. The design of the system is based on the Apache Struts Web
Application Framework .In this way, only one Servlet (FrontController) is needed where all requests go
854
to. The Servlet can analyze every request and execute the action that corresponds to the request. The
Web application can be implemented with Java technology.
JADE (Multi-agent Framework)
Supplier 1
Browser
Web page
Http
Web Server
Web Application
Multi-agent
Shopping System
Supplier 2
Framework Struts
Supplier n
Register and Consult
AMS
DF
White
Pages
Fig. 4.
Yellow
Pages
Framework of Web application
4 Conclusion
To automate shopping process on Internet and help customers to purchase the desirable commodities
from the websites quickly, an agent-based shopping system is proposed in this paper. In the system, we
have integrated the knowledge-based decision-making method and preference acquiring method to make
the best possible commodity for customers. A prototype of the system for used car shopping is
implemented using the Java Agent Development Framework (JADE). The result shows that the system
performs efficiently. We are confident that our system will help customers save enormous time for
Internet shopping. It’s suitable to design personalized service for e-commerce. Our future work will be
focused on (1) developing a more intelligent approach for preference mining in order to trace the
customer’s shopping behaviors more actively; (2) designing a more complete ontology that comprises
more knowledge should be created. The application should be able to take into account and to use all the
knowledge contained in the ontology, such as axioms, attributes etc.
References
[1] Guttman R.H. Agent-mediated electronic commerce: a survey. Knowledge Engineering Review, Vol
13,No.2,pp147-159.
[2] Russell, S., Norvig, P. Artificial Intelligence, A Modern Approach [M]. Prentice Hall Press, 1995.
[3] Judge, P.C. Firefly: The Web Site that Has Mad Ave. buzzing, Business Week online, Available in:
http://www.businessweek.com/1996/41/b349690.htm
[4] Guttman R.H, Maes,P.Agent-Mediated Integrative Negotiation for Retail Electronic Commerce.
Lecture Note in Artificial Intelligence 1571, Agent Mediated Electronic Commerce, pp70-90.
[5] Guttman R.H. Merchant Differentation through Integrative Negotiation in Agent-mediated
Electronic Commerce[M]. Media Art and Sciences, MIT Press, 1998.
[6] The JADE Project Home Page, http://sharon.cselt.it/projects/jade.
[7] Balabanovic,M., Shoham, Y.Content-based collaborative recommendation. Communication of the
ACM, Vol. 40,No.3, pp66-72.
855