A Review of Negotiation Agents in e-commerce by ijcsis

VIEWS: 237 PAGES: 10

									                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                         Vol. 8, No. 3, 2010

       A Review of Negotiation Agents in e-commerce
             Sahar Ebadi*                                Md. Nasir Sulaiman                             Masrah Azrifah Azmi Murad
Department of Information System,                 Department of Information System,                   Department of Information System,
  Faculty of Computer Science and                   Faculty of Computer Science and                     Faculty of Computer Science and
Information Technology, University                Information Technology, University                  Information Technology, University
 Putra Malaysia, Serdang, Malaysia                 Putra Malaysia, Serdang, Malaysia                   Putra Malaysia, Serdang, Malaysia
       Sah_ebadi@yahoo.com                             nasir@fsktm.upm.edu.my                             masrah.azrifah@gmail.com

Abstract—With the growth of World Wide Web and the                         characteristics and concepts used in this field. Then we
increasing human demand on online trading, there is a pressing             explicate the most popular methods in optimizing negotiation
need for complex systems that are capable of handling the client           agent models. Related prominent techniques will be explained
needs in e-commerce. In recent years, numbers of Multi Agent               and some exemplar stand out models will be interpreted.
System (MAS) developers arise to fulfill this mission by
performing a huge number of studies on agent negotiation                       This paper presents two novel classifications, namely
systems. However, far too little attention has been paid to                classification according to the type of agents, and classification
provide a rich review as a repository for developers to distinguish        according to the attitude of agents; in addition to the
the aspect and scope of MAS. The purpose of this paper is to do            Wooldridge’s popular classification according to the number of
a review of progressing agent negotiation technology in e-                 agents involved in negotiation. These classifications help
commerce. In order to achieve our aim we propose different                 developers and researchers to have a better understanding of
classification schemata and interpret different models according           aspects and scope of agent negotiation systems. These features
to the proposed classifications. Popular methods for optimizing            can facilitate future developments on negotiation mechanism in
negotiation agents have been introduced and the effect of relative         MAS. Finally, we will trace the progress of negotiation
techniques has been analyzed. The result of analysis shows that            systems generations by analyzing the most prominent works
genetic algorithm is the most effective learning technique in              during the last decade, from early 1996 to late 2009. In this
optimizing negotiation models. Moreover, we interpret the most
                                                                           work, we will interpret the whole negotiation agent model
prominent negotiation models according to the main parameters
                                                                           according to four important characteristics of the system so-
on which any negotiation agent model depends. The result of
these analysis supplies a resource of differentiating competing
                                                                           called negotiation protocol, negotiation strategy, agent
alternatives for the area of negotiation agent’s models to exploit.        characteristics and negotiation setting. In addition, the most
Finally, a range of open issues and future challenges are                  effective learning technique will be verified through
highlighted.                                                               interpreting the final utility of exemplar models. The remainder
                                                                           of this article is structured as follows. Section 2 presents
KEYWORDS-COMPONENT; ARTIFICIAL INTELLIGENCE; AGENT; MULTI-                 essential characteristics and concepts in the scope of our work.
AGENT SYSTEM; NEGOTIATION; E-COMMERCE                                      Section 3 presents popular methods of optimizing negotiation
                                                                           models and a brief description of the relative techniques. In
                           I.   INTRODUCTION                               section 4, proposed classifications are discussed. Section 5
    With the rapid growth of the World Wide Web and huge                   draws together the two later strands. It discusses some
demand of online trading every day, there is a need for                    exemplar negotiation agent-based applications and highlights
complex systems that are capable of addressing the online                  new direction to the future works. Section 6 presents the
trading needs of human. Such systems must be capable of                    conclusion of this paper.
establishing communications, making decisions and handling
customer’s requirements. Many researchers in Multi Agent                               II.   CONCEPTS AND CHARACTERISTICS
System (MAS) and agent negotiation systems succeed to fulfill                  As mentioned earlier, one of the difficulties in agent
these obligations on e-commerce.                                           negotiation systems is the lack of universally accepted
                                                                           definitions. In order to draw a clear picture of concepts and
    As the number of research conducted on agent negotiation               characteristics of negotiation agents, it is important to recap
development has rapidly increased, the need of conducting a                some key concepts and definitions which are accepted by some
comprehensive review on negotiation agent in e-commerce has                experts in this field.
increased. So far, however, there has been little review about
agent negotiation models specifically in e-commerce. The                      Agent: an agent is a computer system that is situated in
purpose of this paper is to review studies conducted on                    some environment, and is capable of autonomous action in this
negotiation agent systems in e-commerce (online trading). In               environment in order to meet its designed objective [1].
general, lack of universally accepted definitions in negotiation           According to Wooldridge and Jennings [1], the term agent is
agent systems is one of the difficulties in this area. Therefore,          most generally used to denote a hardware or (more usually)
we initially clarify the negotiation agent system’s

    * Responsible author

                                                                      11                               http://sites.google.com/site/ijcsis/
                                                                                                       ISSN 1947-5500
                                                            (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                      Vol. 8, No. 3, 2010
software-based computer system that enjoys the following                     III.   NEGOTIATION AGENT’S CLASSIFICATION
properties:                                                                 Negotiation Agents can be categorized by severed
                                                                        orthogonal dimensions. In this work, we present three
        autonomy: agents operate without the direct                    classifications. The first one is a popular classification
         intervention of humans or others, and have some kind           introduced by Wooldridge [5].        The second and third
         of control over their actions and internal state [2]           classifications are proposed by this work.           Proposed
                                                                        classifications schemata are done according to i) the types of
                                                                        agents involved in negotiation and ii) the attitudes of agents
        social ability: agents interact with other agents (and         involved in negotiation.
         possibly humans) via some kind of agent-
         communication language [3]                                     A.       Number of Agents Involved in the Negotiation
                                                                            This category is one of the most clear and popular
                                                                        categories which are divided into three groups named one-to-
        reactivity: agents perceive their environment, (which          one negotiation, one-to-many negotiation and many-to-many
         may be the physical world, a user via a graphical user         negotiation.
         interface, a collection of other agents, the
         INTERNET, or all of these combined), and respond in               One-to-one negotiation is suitable in situations where one
         a timely fashion to changes                                    agent is negotiating with another agent (e.g. Kasbah model
                                                                        proposed by Anthony et al. [7] and Ryszard Kowalczyk [8]) in
                                                                        case where one agent is involved in the negotiating with other
        pro-activness: agents do not simply act in response to         agent. This is a simple but basic kind of negotiation on e-
         their environment, they are able to exhibit goal-              commerce.
         directed behavior by taking the initiative.                        One-to-many negotiation is when one agent negotiates with
    In most of the real world problems, agents need to interact         several agents as its opponents. This kind of negotiation, in
with other agents to achieve their objectives. Many problem             fact, is originated from several combinatorial one-to-one
cases are innately multi party or social such as negotiation            negotiations.     A practical example of a one-to-many
scenarios. Many are more complex to be solved by an agent               negotiation is auction system where several bidders participate
(e.g. monitoring and performing an electronic marketplace). In          in an auction at the same time. Many researchers [9, 10] on
these cases, multi agent systems are designed to address these          online trading propose their models based on this classification.
issues.                                                                     Many-to-many negotiation as described by Wooldridge [5],
    Multi Agent System: generally Multi Agent Systems (MAS)             is when many agents negotiate with many other agents
refers to such a system that many intelligent agents involved           simultaneously. Jiangbo many-to-many negotiation framework
interact with each other in a selfish or cooperative manner. In         is one of the best examples of this category [11-13].
MAS, agents are autonomous entities and can cooperate to
reach a common goal (e.g. an ant colony) or just follow their
                                                                        B.        Type of Agents Involved in Negotiation
own preferences (as in the free market economy).
                                                                            We believe finding the suitable classification of opponent
    According to Sycara [4], “the characteristics of MASs are           agents type will help agents to choose the best possible tactics
that (1) each agent has incomplete information or capabilities          to deal with their opponents over a particular good. We think
for solving the problem and, thus, has a limited viewpoint; (2)         this classification will decrease the negotiation cost by
there is no system global control; (3) data are decentralized;          improving the accuracy of agent’s beliefs. Many researchers
and (4) computation is asynchronous.”                                   [10, 14, 15] have discussed types of agents involved in a
                                                                        negotiation. According to the variety of application different
    Since trading domains are often bounded by limited                  concepts are offered. Nguyen [10] divided agents involved in
resources and abilities, negotiation has become an essential            negotiation into two groups according to the amount of the
activity in e-commerce applications.                                    concede which they are willing to give. The first type is
    Negotiation: negotiation is a process in which two or more          conceder and the second is non-conceder. The conceder is
parties with different criteria, constraints, and preferences,          referred to the agents that are willing to concede with the aim
jointly reach to an agreement on the terms of a transaction [5].        of selling. In contrast, non-conceder agents are those who just
                                                                        deal with a situation where there is some amount of benefit.
    Negotiation Agents : according to Raymond [6], the notion           Bayesian classification was employed to identify the probable
of agency can be applied to build robust, fast and optimal              types of agents. Then, agents try to modify appropriate
architecture for automated negotiation systems within a group           strategies according to the type of the opponent agents.
of software agents communicating and autonomously making
decisions on behalf of their human users .                                 A new direction in this area is to apply some sense of
                                                                        machine learning techniques to increase accuracy of the
    After a clear understanding of what agent-based concepts            opponent agent classification. Therefore, agent is able of
are, it is necessary to emphasize the key concepts on                   making a better deal that improves the performance of the
optimizing negotiation agent systems on e-commerce.                     model.
                                                                           Ros and Sierra [16] defined 5 types of agents that have been
                                                                        experienced by the model and finally evaluated by utility

                                                                   12                              http://sites.google.com/site/ijcsis/
                                                                                                   ISSN 1947-5500
                                                               (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                         Vol. 8, No. 3, 2010
product and utility difference. These 5 types of agents so-                according to the agent’s behavior as proposed by Ikpeme et al.
called NegoTo agent, Random agent, Alternate agent, TOagent                [15].
and Nego agent are differentiated by tactics and behaviors they
obey. These 5 types of agents are explained In a nutshell.                     The proposed classifications will clear the roadmap for
NegoTo agent, before reaching deadlock applies trade-off                   developers in establishing new research in the area of
tactic, followed by negoEngin tactic. Random agent chooses                 negotiation agent systems. Once developers define the agent
the next tactic randomly (e.g. negoEngin, trade-off, trade-off             classification schema, automatically dimensions and scopes of
and negoEngin). The above-mentioned agents alter negoEngin                 research area are clarified. Simply, by following previous
and trade-off tactics one at a time. TOagent obeys trade-off               works conducted in the specified class, the gaps and
tactics when utility of new offer is higher than the previous              disadvantages of research area will be highlighted. This will
one, otherwise, the aspiration level is decreased and new offer            assist researchers to develop and maintain innovative research
is proposed. Nego agent only follows negoEngin tactics during              in the area of negotiation agent models.
the negotiation.                                                       IV.         POPULAR METHODS OF OPTIMIZING NEGOTIATION
    We think that applying a decision making model that                 SYSTEMS
discovers the highest utility among all tactics can increase the             Designing an optimized agent negotiation system is one of
performance of the system. In addition, applying the previous           the important issues addressed recently by many researchers
history or knowledge extracted from past experience (e.g. what          [16, 19-23]. In recent years, a great deal of effort has been
happened in the past when we used a specific tactic in the same         devoted toward optimizing negotiation agents[24-27]. Mostly
situation?) in the same situation can help agents decision              this aims at improving agent i) adaptability, ii) intellectuality,
making. This will result in a more accurate choice of tactics or        iii) applied strategies, and iv) gathering information.
type of agents to negotiate and end in a higher chance of                  A.        Adaptability
reaching agreement.                                                            Negotiation agents are situated in open environment where
                                                                           new agents may come and some agents may leave the
                                                                           environment. Agents are characterized by deadlines, volatile
C.         Attitude of Agents Involved in Negotiation                      preferences, and incomplete information. In such situations,
     Internet is populated by many agents coming from different            agents must survive by changing their strategies, preferences
sources with different attitudes and goals. Some researchers               and even learning opponent’s preferences and behaviors. This
mentioned that, although agents can be categorized by their                attempt to change agent’s behaviors, preferences, and strategies
behaviors, they can also be categorized according to their                 toward reaching an agreement is called adaptability. This
attitudes toward their goals. In a clear word, agent’s attitude            method aims to find the highest satisfactory offer for agents as
defines how an agent selects beneficial opponent in a given                well as being acceptable for opponent agent. Many researchers
different situation[17]. Many researchers are working on this              [6, 27, 28] applied this methods for optimizing negotiation
area in order to reach to a better efficiency in their models [10,         systems, as it is effective and necessary to assist the autonomy
15, 17, 18] . In Ikpeme proposed model [15] agents are divided             aspect of agents. Zhang et al. [27] proposed a new adaptive
according to their social attitudes into three different groups so-        negotiation strategy which assists negotiation models by
called helpful, reciproactive, and selfish. Helpful agents are             enhancing the adaptability of agents. In this scenario, an agent
those who are willing to help. This group benefits in the                  uses adaptive strategy to estimate the strategies and preferences
homogenous groups of helpful agents. Reciproactive agents                  that opponent agents used in the last few offers in negotiation.
evaluate the request and will accept opponent’s request if the             So, an agent can choose appropriate negotiation factors to
agent meets a balance of cost and saving, otherwise they reject            adjust its strategy. Even after choosing the strategy, there is a
the request. This group has the best performance when situated             chance to change the parameters of chosen strategy to adapt
in an open group of agents. Selfish agents never help the other            dynamically to the strategy of another agent. Also, Raymond
agents. So, they always benefit when situated among helpful                [6] proposed a negotiation mechanism in which an agent can
agents but rarely benefit from reciproactive agents. Finally, the          adapt itself by changing its preferences and behavior using
results show that in an open environment success rate is many              dynamic models. This learning model is applied by using
times better for reciproactive agents than selfish agents.                 Genetic Algorithm (GA) in process of decision making, so that
    In the work presented by Jaesuk et al. [17], agents attitude           agent can gradually acquire appropriate negotiation knowledge
defines the priority that an agent places on various choices it            based on its negotiation history with that opponent. Thus,
may have regarding member selection. So, agent’s attitude is               Raymond negotiation model aims to reach the same property
(1) toward rewards or (2) toward risk. Agent attitude toward               but by using machine learning techniques.
reward is the agent’s point of view toward finding the best
                                                                           B.        Intellectuality
opponent. The attitude considers the opponent’s quality of
                                                                               As intellectual capability is one of the substantial
service. Agent attitude toward risk is agent sensitivity to the
                                                                           characteristics in agent technology, learning appears as one of
possible risk of opponent agent (which depends on unreliability
                                                                           the most important and effective methods in improving
and unavailability of agents). The result of the research shows
                                                                           negotiation agents. Recently, many researchers enriched their
that agents with strong attitude toward risk are more beneficial
                                                                           negotiation agents and models by using different kinds of
when there is a higher chance of failing jobs due to tight time
                                                                           learning algorithms or machine learning techniques [6, 9, 26,
or low availability and reliability of opponent. But, agents with
                                                                           29-32]. Former attempts in this area were applied by using
the high attitude toward quality situated in a case where time is
                                                                           some sort of machine learning techniques (e.g. GA, fuzzy, re-
enough and the penalty value is small are able to earn more
                                                                           enforcement learning, simulated annealing and data mining),
benefit. Also, in some cases agent’s classification is done
                                                                           distribution probabilistic analysis or by applying some pre-

                                                                      13                              http://sites.google.com/site/ijcsis/
                                                                                                      ISSN 1947-5500
                                                                                 (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                           Vol. 8, No. 3, 2010
programmed strategies. In Intelligent Trading Agents(ITA)                                 Tow-Step clustering algorithm is used to cluster buyers
model [9] four different types of strategies were introduced to                           according to agent’s characteristics. After that, a rule-based
help agent’s to choose the next action to be taken. E-                                    modeling technique, C5.0 classification algorithm [26] is used
Commerce Negotiation (E_CN) [10] also presents complex                                    to extract the consumption pattern of each cluster population.
negotiation tactics combined with a sort of distribution                                  Finally, Apriory algorithm [26] is used to discover association
probabilistic analysis to predict the opponent’s type which will                          between buyer details and purchases. A future direction for
end to choose the best possible strategy for the next round.                              this research is to employ some sense of artificial intelligence
Other researchers tried to employ machine learning techniques                             to the seller agent which will result in a higher overall
directly to the agent’s decision making engine. For example,                              performance of the system.
Raymond [6] applied a genetic algorithm on the space of
possible offer for agent A to propose to agent B. Fig. 1
represents an example of how agent offer encodes to the                                   C.        Applied Strategies
chromosomes and draws a tow-point crossover operation                                         The outcome of negotiation depends on several parameters
according to Raymond Lau[6].                                                              including agent’s strategy. Strategies are divided into three
                                                                                          groups: (1) strategies which depend on time called time-
                                                                                          dependent strategies, (2) strategies which depend on agent’s
                                                                                          behavior, called behavior-dependent strategies and (3)
         Double Crossover                                                                 strategies which depend on how a specific resource is
                                                                                          consumed. Recently there have been huge amounts of research
                                                                                          on agent’s strategies [9, 10, 26, 27, 29].

          Parent Offer1                  20-40           10-20         10-30                  In ITA model [9], some time-dependent strategies were
                                                                                          proposed, namely Desperate, Patient, Optimized patient and
                                          Price           Quality       Time
                                                                                          strategy Manipulation. Desperate strategy accepts the first

          Parent Offer2                  30-50           20-50         10-20
                                                                                          acceptable offer that is suitable. In this strategy agent aims to
                                                                                          reach a deal as soon as possible. Patient strategy waits until all
                                          Price           Quality         Time            negotiation threads reach a deal and then choose the best offer.
                                                  Crossover         Crossover
                                                   Point 1           Point 2              This strategy guarantees the best possible deal but does not
                                                                                          consider time constraint which is one of the most important

          Child Offer                    30-50            10-20        10-30              factors in real market places. In Optimized patient strategy, the
                                          Price           Quality       Time
                                                                                          outcome of one sub-negotiation affects the performance of
                                                                                          other sub-negotiations. Manipulation is a combination of
                                                                                          above-mentioned strategies. A drawback of such strategies is
                                                                                          the lack of adaptability as they are preprogrammed.

                Figure 1.             Encoding Candidate Offers[6]                            E-CN model [10] used time-dependent strategies called
                                                                                          Conceder, Linear and Tough. Conceder strategy means that an
    In this scenario agent A will generate a random pool of                               agent quickly lowers its preference values until it reaches its
possible offer to propose to agent B. Then GA is applied on                               minimum reserved value. Linear strategy is when agent drops
mating pool which is using three standard genetic operators:                              its reserved values but in a gradual manner. Tough strategy
cloning, crossover, mutation. The outcome is a list of                                    deals in a tough manner, meaning that it keeps its values until
generated possible offers which are ranked by their utility                               agent is close to deadline then suddenly drops the values to its
values. The first offer in the list is the offer to be proposed to                        reserved values. However, these sorts of strategies are sub-
agent B by agent A. Proposed GA-based negotiation agents                                  optimal in which using a sense of learning can improve the
assist agents by learning their opponent’s preferences. This                              efficiency and robustness of system.
improves the speed of the search in finding the best possible                                 Isabel et al. [26] proposed two types of behavior-dependent
offer for both agents. Although this learning ability increases                           strategies named Composed Goal Directed (CGD) and
the overall utility rate to 10.3%, still there are some open issues                       Adaptive Derivate Following (ADF). CGD is based on two
to increase the performance of the system. One of the open                                objectives which should follow sequentially. The first objective
issues is to acquire knowledge extracted from the recorded                                is to be sure that all needed goods are sold or purchased. The
history or a trusted third party in order to setup the GA                                 second tries to reduce the pay off of the deal or increase the
initialization accurately (e.g. possible range of value for genes).                       benefit. ADF is based on the revenue earned in the previous
    Guo et al. [32] suggested an algorithm to extract knowledge                           period as a result of changes in the price. If the change of price
from a user and then inject it into the solution population.                              by the last period produced more revenue than the previous
Then, a simulated annealing was applied to render the solution                            period, then the strategy takes a similar action otherwise it will
to make sure that the best possible offer will be proposed.                               take a counter action.
    Isabel et al. [26] proposed a multi-agent market simulator.                               Although, time-dependent strategies seem simple, with
In this market, agents have the possibility to negotiate through                          concern to the time dependent of negotiation agent, they have a
the pool which is regulated by a market operator (market                                  significant effect on system’s outcome. A combination of
administrator). Three data mining techniques are proposed to                              alternatives of different types of tactics were proposed in this
mine the administrator’s transaction history, which contains all                          area [16, 23].
previous interactions and transactions among agents. The

                                                                                     14                               http://sites.google.com/site/ijcsis/
                                                                                                                      ISSN 1947-5500
                                                                              (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                        Vol. 8, No. 3, 2010
D.        Gathering Information                                                             In that second step called “applying external knowledge”,
    The lack of information about environment and opponent                              knowledge gathered from outside is obtained directly from user
agents is one of the major issues on negotiation systems [18,                           input or derived from user observation. The base solution is
26, 32-35] since these information assist the agents to choose                          now modified by injecting external knowledge into the solution
suitable agents and strategies. Such information can be                                 population. The knowledge will be accepted based on
gathered via agent’s recorded negotiation histories or via                              correctness of the knowledge and a certain probability [32] .
trusted third party agents and referral systems. The referral                           Guo et al. [32] showed that taking this knowledge into account
mechanism allows agents to find their required resources if                             while generating solution population, has a significant effect on
there is any agent with the required expertise close to the                             learning user preferences in multi-attribute negotiation.
location of the neighboring agents [35]. Many researchers [18,
                                                                                            Choo et al. [33] also presented a form of optimizing
26, 32-36] have focused on this issue with the aim to improve
                                                                                        negotiation agents by employing genetic algorithms. This
their negotiation agent models. Guo [32] in his proposed
                                                                                        model attempts to learn the agent opponent’s preferences
algorithm referred to information as an important factor which
                                                                                        according to the history of the counter offers based upon the
assists suggested model in learning multi-attribute utility
                                                                                        stochastic approximation. One of the further discussions in the
function. Gathering information is mentioned as the second
                                                                                        agent models is employing trust on negotiation agents
step of the Guo algorithm Fig. 2.
                                                                                             Trust can help agents to follow their aims by gathering
                                                                                        useful information about their opponent’s preferences and their
                                                                                        attitudes toward their goals. Multi Dimensional Trust (MDT)
             Utility Elicitation Algorithm                                              introduced by Griffiths et al. [37] opened a future direction in
                                                                                        the area of negotiation agent for finding trustworthy opponent.
                  Applying Evolutionary Operations                                      Griffiths et al [37] applied a weighting factor concept which
                         Selection, Crossover and Mutation
                                                                                        enables agents to combine decision factors according to agents
                                                                                        current preferences. Later on, Gujral [38] provided its agents
                                                                                        with a model of recognizing the best trusted third party to
                                                                                        obtaining information from. In that model, agents should
                      Applying External Knowledge                                       consider the trustworthiness of the potential opponents in order
                      Knowledge Acquisition and Integration                             to maximize the agent’s rewards, inasmuch as the more reliable
                                                                                        the trusty agent is, the higher the chance of reaching agreement.
                                                                                             There are some other important factors in multi agent
                                                                                        negotiation systems which, taken into account, can affect the
                            Apply Local Search
                    Local Refinment with Simulated Anealing
                                                                                        outcome of negotiation’s systems. These include the number
                                                                                        and type of issues considered in agent services [16, 39], agent
                                                                                        attitude [17], one-sided or two-sided commitment [11, 40],
                                                                                        bilateral [39] or multilateral [6] negotiation.
                 No                                      yes
                             Stop Criteria met?                End
                                                                                        The standout research and relative optimization techniques and
                                                                                        methods of negotiation agent models are summarized in table 1

                 Figure 2. Utility elicitation algorithm[32]

     Optimization Me thods                Te chnique s                     Exe mpular de s cription                    Stand out Re s e arche s

                                          Referral Systems                 Asking information from neighbors           Ebadi et.al. (2008)
     Gathe ring information               Trust                            Based on trust worthy of opponent           Griffiths (2005),Gujral et.al.(2007)
                                          History                          based on recorded experienced               Choo et.al.(2009), Guo et.al.(2003)

                                          adaptive learning techniques     learnin adaptive factor during trading      Raymond (2009),Magda et.al.(2009)
     Adaptability                         Flexible models                  overcoming pre-programmed tactics           Zang et.al(2007),chung-wei(2008)
                                          Dynamic models                   Dynamic methods and programming             Raymond (2009)

                                          Marchine learning techniques:    Genetic Algorithm                           Choo et.al.(2009), Magda et.al.(2009)
     Inte lle ctuality                                                     Bayesian Learning                           Duong e.al.(2004),Choo et.al(2009)
                                                                           Reinforcement Learning                      Sen et.al.(2007)
                                                                           Fuzzy Logic                                 Cheng et.al.(2005)
                                                                           Evolutionary Learning                       Raymond(2009)
                                          Data mining techniques           Apriory algo/Classification(C5)Algo         Isabel et.al.(2008)
                                          Others                           Simulated Annealing                         Isabel et.al.(2008)

                                          Time dependant tactics           Boulware/Conceder/Linear                    Iyad et.al.(2002),Doung et.al.(2004)
                                                                           Relative tit-for -tat/Random absolut tit-
     Applie d Strate gie s                imitative tactics                for-tat/Averaged tit-for-tat                Isabel et.al.(2008)
                                          Resource dependant tactics       Dynamic deadline/Resource estimation        not common

                                                                                   15                                     http://sites.google.com/site/ijcsis/
                                                                                                                          ISSN 1947-5500
                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                     Vol. 8, No. 3, 2010
                    V.    APPLICATIONS                                 desperate, patient and optimal patient. Optimal patient helps
                                                                       agents to assist agent’s autonomous behavior in dynamic
    After specifying the domain of classification and                  environments. In this case, preferences of agents will be
optimization of negotiation agents, some outstanding works             marked by weighting factor which represents the degree of
carried out during recent years will be reviewed. Following the        importance of every issue.
new generation of the models we will track the progressing
flow of agent generations. In addition, the whole negotiation              Weighting factor was firstly introduced by Griffiths [37].
systems proposed in the models will be analyzed to elaborate           Later on, this concept under the term of Multi-dimensionality
on advantages and drawbacks of these systems. These systems            [38] was used by many agent researchers [43, 44] to evaluate
are analyzed in terms of desirable negotiation protocol,               their standard measurements and finding the more appropriate
negotiation strategy, agent characteristics and negotiation            deal or opponent. However, ITA’s strategies help agent to gain
setting.                                                               higher performance but still the act of agent is bounded by
                                                                       premier choice of strategies before each round of negotiation.
  1) Desirable negotiation protocol represents the rules that          This will reduce the adaptability of negotiation agent.
govern the interaction between negotiators [22]. According to
Nguyen and Jennings.[41], desirable properties for a
negotiation protocol include pareto efficiency and Guarantee of
                                                                        Dynamic Environment
  2) Negotiation strategy is the arrangement of a series of                        Buyer agent
actions that the agents plan to take through the negotiation.
Negotiation tactics specify whether it is time dependent,
                                                                                                Sub-buyer 1                        Seller 1
resource dependant or imitative. In some cases, negotiation
tactics are a combination of aforementioned tactics proposed by
models.                                                                    Coordinator          Sub-buyer 2                        Seller 2
  3) Agent characteristics specify agent’s knowledge and
experience, learning and adaptation capabilities.
  4) Negotiation setting deals with factors which are related                                   Sub-buyer n                        Seller n
to problem domain. It includes number of negotiation issues
and number of parties involved.

   The analysis will deal with the evolution that the primary                 Figure 3. System architecture adapted from ITA system [9]
Kasbah model [7] underwent from 1996 to 2009, leading to
GAMA model (table2)
    Kasbah model [7] is a simple one-to-one negotiation                    The new version of ITA called e-CN was proposed by
framework. Its application is on e-commerce where the agent            Nguen and Jennings [10]. This method uses several number of
technology will meet the web-based systems and try to                  agent services’ issues considered in negotiation. Also, a factor
overcome the need of online trading by applying some                   is introduced as “probability of agent distribution” which
autonomy on trading. Kasbah model is single issue and                  represents the probability of allocating types of agent in the
considers price as the most important issue in negotiation.            environment. Nguyen [10] believes that agents can be divided
Therefore, agents are searching to make a deal with appropriate        by their behavior into two groups of agents called Conceder
price even before they meet their deadline. Lack of adaptation         and Non-conceder. Conceder stands for agents who are willing
and intellectuality is obvious draw-back of Kasbah’s                   to concede in order to reach a deal while Non-conceder stands
negotiation agents which is overcome by defining new versions          for agents who are not willing to sell, otherwise there is a
of Intelligent Trading Agents such as ITA and e-CN.                    special amount of benefit for them, so they act in a tough
    ITA [9] is a one-to-many negotiation framework which is            manner.
an improved version of negotiation systems in terms of number             In e-CN model, agents choose their strategies based on a
of issues considered in negotiation and in terms of                    method of predicting the expected utility of chosen strategy
communicating as it follows bilateral negotiation. In bilateral        considering the current situation. This Expected utility will be
negotiation, agents have the ability to send offer and counter-        evaluated according to 3 important probability factors:
offer in both ways. ITA presents new system architecture               probability of agent’s distribution, possibility of reaching
represented in Fig. 3.                                                 agreement and average utility value if agent reaches the
    ITA buyer agent includes two components, namely                    agreement.
coordinator and sub-buyers. Buyer agent will establish several             The disadvantage of proposed concession strategies in e-
one-to-one negotiations between sub-buyers of buyer agent and          CN and ITA is that in every round of offering new proposal,
sellers. In this work, agent preferences are represented as a          agents are conceding while there is a possibility to find a
constraint satisfaction problem as described by Vipin [42].            mutually acceptable proposal with the same utility level but
Moreover, this model proposes different sort of strategies like        different values of the issues. Future research should be done

                                                                  16                                 http://sites.google.com/site/ijcsis/
                                                                                                     ISSN 1947-5500
                                                              (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                        Vol. 8, No. 3, 2010
to investigate a sort of similarity function that reveals the                 Raymond [22] performed a similar series of experiments in
negotiable issues with higher importance. Finding important               terms of models characteristics. Raymond negotiation agents
negotiable issues assists agents to reach to an agreement using           are enhanced as they are promising in supporting real-world e-
the slightest possible amendments. We believe this will result            market places environment. This application sustains multi-
in higher optimization and also faster deals in terms of time             party, multi-issue, many-to-many negotiation which are based
consuming. Aforementioned future developments will result in              on parallel and distributed decision making model. Moreover,
decreasing negotiation cost.                                              Raymond [22] introduced his novel genetic algorithm in this
                                                                          experiment. The final result of the experiment showed that an
    Although these models support bilateral multiple                      evolutionary negotiation agent guarantees pareto optimal
concurrent negotiation, there is still much more need to assist           solutions underneath dynamic negotiation situation, for
the agent decision making methods. Helping agent to predict               example in the incidence of time limitations.
its opponent next action or finding the opponent preferred
issues is another future work. As we discussed in section IV                  However, as we mentioned in section IV D, embedding a
this objective could be achieved by enhancing the                         sort of strategies could enhance the success rate of the two
intellectuality of agents. Referring to table 1 there are many            aforementioned negotiation systems.
possible techniques to investigate as future studies. An
appropriate learning technique could lead this model to find the              Magda [45] introduced an agent mediated shopping system
pareto optimal solution.                                                  called Genetic Algorithm driven Multi-Agents (GAMA).
                                                                          GAMA is a multi issue bilateral negotiation system enhanced
    Cheng [23] proposed a heuristic, multiple issue, one-to-              by learning ability. GAMA studies the effect of participating
many model for negotiations in a third-party-driven e-market              opponent agent’s preferences in decision making of agents. In
place. These negotiation agents employ trade-off tactics using            order to do that, one of the offers from the opponent agent’s
fuzzy inference systems to generate new offer in each round.              previous offers (or list of offers) is chosen as one of the parents
Trade-off tactics are navigated from time-dependent and                   and the other parent is chosen from agent’s own preferable
imitative tactics.                                                        proposals. Then, the mutated offspring is generated. The new
                                                                          generated offspring is a potential satisfactory offer as it is
    Although, Cheng’s proposed model is pareto-efficient and              mutually acceptable for both agents. Experimental results
highly adaptive revealing the importance level of issues to the           demonstrated that GAMA achieved to a higher satisfaction rate
other agents, it violates the privacy of information. These sorts         while reaching to the higher numbers of the deal in comparison
of assumptions are highly inappropriate since they are hardly             with traditional GA methods.
acceptable in the real world environments such as e-markets.
                                                                              One of the advantages of this work considering these parent
    A negotiation meta strategy for one-to-one bilateral                  selections is increasing the adaptability of the system, since
negotiation was proposed by Ros and Sierra[16]. Meta strategy
                                                                          every change of opponent agent’s preferences effects the
is a combinatorial sequence of concession and trade-off tactics           decision making of agents. In addition, proposing mutually
which will try to keep the aspiration level, otherwise there is no        admissible offer causes to reach an agreement in fewer rounds
possible offer by the current aspiration level. Combining                 of negotiation, thus reducing the cost of negotiation. In future
tactics allows agents to outperform better in different situation         investigations, it might be possible to experiment this model
which fulfills the adaptive capability of agents. In addition,            under qualitative issues as an alternative of quantitative issues.
detection of opponent agent preferences helps agents to                   This could result in higher optimization and also faster deals in
propose mutually acceptable offers. As a consequence, the                 terms of time consumption.
system’s final success rate increases.
                                                                             In order to have a better understanding on the overview of
    Although this model is placed in a satisfactory level of              applications discussed, table2 illustrates the general
adaptability, it is recommended that further research be                  characteristics of the applications discussed.
undertaken in order to learn the opponent agent’s type. Finding
the correct classification for type of agent could increase the
chance of reaching agreement.

    Choo et al. [19] conducted a research on one-to-many
bilateral negotiation with multiple issues (quantitative issues).
The system architecture of the study was based on ITA’s
system architecture. They investigated two different machine
learning approaches genetic algorithm and Bayesian learning
called GA improved-ITA and Bayesian improved-ITA. The
result obtained from the final analysis showed that GA-
improved-ITA outperforms, Bayesian-improved-ITA in
maximizing the joint utility and negotiation payoff at the same
time as it increases the negotiation cost.

                                                                     17                               http://sites.google.com/site/ijcsis/
                                                                                                      ISSN 1947-5500
                                                                                 (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                           Vol. 8, No. 3, 2010

Negotiation system characteristics       Kasbah(1996)     ITA(2001)         e-CN(2004)         Cheng et.al.(2005)Ros et.al.(2006) Choo et.al.(2009) Raymond(2009) GAMA(2009)
Cardinality of Negotiation Domain         single-issue    multiple-issue    multiple-issue        multiple-issue   multiple-issue   multiple-issue   multiple-issue multiple-issue
Cardinality of Communication               unknown          bilateral         unknown                bilateral       bilateral        bilateral       mutilateral     bilateral
Number of agents involved in negotiaion one-to-one        one-to-many       one-to-many           one-to-many       one-to-one      one-to-many      many-to-many   one-to-many
Qualitative negotiation value of issues         −               −                 √                      √               √                −                √              −
Quantitative negotiation value of issues        √               √                 √                      √               √                √                √              √
Privacy of model                                √               √                 √                      √               √                √                √              √
Privacy of information                          √               √                 √            not-private.see Des       √                √                √              √
Negotiation tactics includes:
TIme/REsource dependant or IMitative            −        Time dependant      mix(TI+IM)         Trade-off(TI+IM)                         Mix(TI+IM)                   Time dependant Time dependant                                                 −
                                                                                                                                                                     Genetic Algorithm,    Genetic                                               Genetic
Intellectuality                               −                 −          Bayesian learning     Fuzzy inference                                         −
                                                                                                                                                                     Bayesian learning    Algorithm                                             Algorithm
Pareto efficiency                             −                 −                 −              Preto Optimal                                           −                   −          Preto Optimal                                               −
Guarantee success                             −                 √                 √                    −                                                 −                   −                −                                                     −
                                                                                                NegoTo, Random,
                                                                              conceder             aLternate,
Type of agents involved                    unknown          unknown                                                                           unknown                         unknown                          unknown                          unknown
                                                                            non-conceder           TOAgent,
Adaptiveability                          non-adaptive     non-adaptive      semi-adaptive           adaptive                                  adaptive                 non-adaptive                            adaptive                         adaptive

    This table shows that embedding strategies in negotiation                                      Artificial intelligence techniques such as GA [6, 19, 32] ,
agent models can increase the final system outcome. This                                       fuzzy logic [23, 25], simulated annealing [32] ,neural network
enhancement could be in terms of increasing adaptability or                                    [47, 48] and re-enforcement learning [30] have been used
assuring the guarantee of success. As illustrated, most of the                                 broadly to improve agent’s intellectuality in recent years.
models empowered by machine learning techniques are fully                                      According to our review, GA is the most popular learning
adaptive. As discussed in section IV A, adaptability is an                                     technique among others as usually models that employ GA in
important characteristic for negotiation system embedded in                                    negotiation systems reach to a higher final utility than other
open environment. Also, as we can see in e-CN and Cheng et                                     techniques (e.g. [22, 49]) as shown in Fig. 4.
al. models [10, 23] an accurate classification on agent’s type is

                                                                                                                                                                                                                                                    Samuel et al. (2001)
driving the negotiation model to a desirable level of                                                              1.8

                                                                                                                                                                                                                                                                           Lau (2009)
adaptability.                                                                                                      1.6

    In order to reach to the pareto optimal result, agents most be
                                                                                                                                                Nguen et al.(2004)


                                                                                                                                                                                                               Park (2004)

                                                                                                                                                                                                                             Soo et al.(2002)
equipped with an effective learning method or a suitable
                                                                                                                         Choo et al. (2009)

                                                                                                                                                                                         Cheng et al. (2005)
strategy. These cases show that learning ability helps agents to
                                                                                                                                                                        Richter (2010)
                                                                                                   Utility value

predict opponent’s characteristics (e.g. preferences, reservation                                                   1
value, attitude and type) accurately. Therefore, by choosing
the best possible strategy (action) we can reach to a pareto
optimal result.                                                                                                    0.6

    Every negotiation mechanism is desirable to meet some                                                          0.4

important requirements. These include pareto efficiency and                                                        0.2
guarantee of success. The power of applying suitable strategy
is revealed by assuring the guarantee of success only by                                                            0

employing the appropriate strategy.
    We believe tables 1 and 2 provide a good resource for
future developments. Since table 2 highlights the shortcomings
and loopholes of the above-mentioned models (represented by
“-” or “non”), developers can easily investigate future studies                                  Figure 4. Comparison of final utility value of different machine learning
in order to overcome these gaps by the help of the optimization                                                                 methods
methods introduced in table1.
                                                                                                   This is due to the pattern of common problems defined in
   As this study shows, artificial intelligence is an important                                negotiation which match GA technique characteristics, since, in
characteristic of multi agent systems which has been used as a                                 typical negotiation problems, we face a big space of feasible
popular enhancing technique to optimize the final outcome of                                   solution and the goal is to find the best solution. Furthermore,
many negotiation agent based systems. As Jennings [46]                                         specifying the fitness function for population ranking is a
admits” undoubtedly the main contributor to the field of                                       straightforward task as generally it equals to utility function for
autonomous agents is artificial intelligence”.                                                 evaluating offers.

                                                                                       18                                                                             http://sites.google.com/site/ijcsis/
                                                                                                                                                                      ISSN 1947-5500
                                                                           (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                     Vol. 8, No. 3, 2010
                          VI.     CONCLUSION                                            [8] V. B. Ryszard Kowalczyk "On Constraint-Based Reasoning in e-
    In the last few years we have witnessed an incredible                               Negotiation Agents, " Lecture Notes In Computer Science, vol. 2003, pp.31-
growth of multi agent systems on e-commerce. One of the                                 46, 2001
most important driving forces behind MAS research and                                   [9] I. Rahwan, R. Kowalczyk and H. H. Pham, "Intelligent agents for
development is the internet. By increasing the human demand                             automated one-to-many e-commerce negotiation, "Proceedings of the twenty-
on online trading, negotiation has become one of the most                               fifth Australasian conference on Computer science , vol.4, pp.197-203, 2002
important topics in MAS. In this area, many researchers tried                           [10] T. D. Nguyen and N. R. Jennings, "Coordinating Multiple Concurrent
to improve the performance of the negotiation agent’s model.                            Negotiations, "Proceedings of the Third International Joint Conference on
This is achieved by using adaptive methods, employing                                   Autonomous Agents and Multiagent Systems , vol.3, pp.1062-1069, 2004
learning abilities, applying strategic reactions and gathering
                                                                                        [11] J. Dang and M. N. Huhns, "Concurrent Multiple-Issue Negotiation for
information. To sum up, negotiation agents can act more
efficiently when they are empowered with effective methods                              Internet-Based Services, " IEEE Internet Computing, vol. 10, pp. 42-49, 2006
for gathering information which assists the agents in employing                         [12] D. Jiangbo and H. M. N., "Coalition deal negotiation for services, "
strategic actions to reach their goals. In addition, agents must                        Rational, Robust, and Secure Negotiation Mechanisms in Multi-Agent
be adaptive by changing their attitudes and learning their                              Systems, 2005, pp. 67-81, 2005
opponents’ characteristics and preferences to improve the                               [13] J. Dang, J. Huang and M. N. Huhns, "Workflow coordination for
overall performance of the system.                                                      service-oriented multiagent systems, "Proceedings of the 6th international
     In order to improve agent negotiation systems, we need to                          joint conference on Autonomous agents and multiagent systems, pp. 2007
understand the dimensions and range of options in these areas.                          [14] W. Danny, S. Kurt, H. Tom and G. Olivier, "Towards Adaptive Role
To set up the foundation, we have developed a classification                            Selection for Behavior-Based Agents, "Adaptive Agents and Multi-Agent
scheme which is specially aimed at negotiation agent systems                            Systems III, pp. 295-312, 2005
on e-commerce. Negotiation agents can be categorized into                               [15] I. Erete, E. Ferguson and S. Sen, "Learning task-specific trust decisions,
different groups with respect to (1) number of agents involved                          "Proceedings of the 7th international joint conference on Autonomous agents
in negotiation, (2) type of agents involved in negotiation, and                         and multiagent systems , vol.3 pp. 2008
(3) according to negotiation agent attitude involve in                                  [16] R. Ros and C. Sierra, "A negotiation meta strategy combining trade-off
                                                                                        and concession moves, " Autonomous Agents and Multi-Agent Systems, vol.
    This classification system was demonstrated on an assorted                          12, pp. 163-181, 2006
range of outstanding negotiation model and the outcome is                               [17] J. Ahn, D. DeAngelis and S. Barber, "Attitude Driven Team Formation
summarized in table 2. The purpose of this classification was                           using Multi-Dimensional Trust, "Proceedings of the 2007 IEEE/WIC/ACM
to present a complete and systematic source to objectively                              International Conference on Intelligent Agent Technology, pp. 229-235,2007
compare and contrast different negotiation models. Such a                               [18] E. Toktam, P. Maryam and P. Martin, "Partner Selection Mechanisms
classification method is essential for developers as it supplies a
                                                                                        for Agent Cooperation, " Web Intelligence and Intelligent Agent Technology,
resource of differentiating competing alternatives for the area
                                                                                        2008. WI-IAT '08. IEEE/WIC/ACM International Conference on, vol. 3, pp.
of negotiation agent’s models to exploit.
                                                                                        554-557, 2008
                                                                                        [19] M. N. S. a. M. H. S. S.C. Ng, "Machine learning approach in optimizing
                     VII.          REFERENCES                                           negotiation agents for e-commerce, " Information Technology Journal, vol. 8,
                                                                                        pp. 801-810, 2009
[1] M. Wooldridge and N. R. Jennings, "Intelligent Agents: Theory and                   [20] L. Alessio, W. Michael and J. R. Nicholas, "A Classification Scheme for
Practice, " Knowledge Engineering Review, vol. 10, pp. 115–152, October                 Negotiation in Electronic Commerce, " Group Decision and Negotiation, vol.
1995.                                                                                   12, pp. 31-56, 2003
[2] C. Castelfranchi, "Guarantees for autonomy in cognitive agent                       [21] P. Faratin, Carles Sierra and N. R. Jennings, "Negotiation Decision
architecture, " Proceedings of the workshop on agent theories, architectures,           Functions for Automated Agents, " Elsevier Sience, vol.24 pp.159-189, 1997
and languages on Intelligent agents, vol. 890, pp. 56-70, 1995                          [22] R. Y. K. Lau, "An Evolutionary Approach for Intelligent Negotiation
[3] S. P. K. Michael R. Genesereth "Software Agents, " Communications of                Agents in e-Marketplaces, " Intel. Agents in the Evol. of Web & Appl, vol.
the ACM, vol.37, pp. 48–53, 1994                                                        167, pp. 279–301, 2009
[4] K. P. Sycara, "Multiagent Systems, " AI Magazine vol. 19(2), pp. 79-92,             [23] C.-B. Cheng, C.-C. h. Chan and K.-C. Lin, "Intelligent agents for e-
1998.                                                                                   marketplace: Negotiation with issue trade-offs by fuzzy inference systems, "
[5] M. Woolridge and M. J. Wooldridge, An introduction to multi agent                   Decis. Support Syst., vol. 42, pp. 626-638, 2005
systems,10 ed., John Wiley & Sons, Inc. New York, NY, USA 2001,                         [24] M. A. Lopez-Carmona, I. Marsa-Maestre, J. R. Velasco and E. d. l. Hoz,
[6] R. Y. K. Lau, "Adaptive negotiation agents for e-business, "Proceedings             "Using Clustering Techniques to Improve Fuzzy Constraint Based Automated
of the 7th international conference on Electronic commerce, pp. 2005                    Purchase Negotiations, " Springer, Advances in Agent-Based Complex
[7] C. Anthony and M. Pattie, "Kasbah: An agent marketplace for buying                  Automated Negotiations, pp. 89–117, 2009
and selling goods, " In Proceedings of the first international Conference on the        [25] J. Richter, "Multistage Fuzzy Decision Making in Bilateral Agent
Practical Application of Intelligent Agents and Multi-Agent Technology, pp.             Negotiation, " 3sd PHD symposium, pp. 71-73, 2010

                                                                                   19                                    http://sites.google.com/site/ijcsis/
                                                                                                                         ISSN 1947-5500
                                                                          (IJCSIS) International Journal of Computer Science and Information Security,
                                                                                                                                    Vol. 8, No. 3, 2010
[26] I. Pra, M. Jo, o. Viamonte, Z. Vale and C. Ramos, "Agent-based                    [40] T. D. Nguyena and N. R. Jennings, "Managing commitments in multiple
simulation of electronic marketplaces with decision support, "Proceedings of           concurrent negotiations, " Electronic Commerce Research and Applications,
the 2008 ACM symposium on Applied computing, pp. 2008                                  vol. 4, pp. 362-376, 2006
[27] Z. Junyan, T. Jiang and D. Gang, "Agent-based multi-factors adaptive              [41] P. F. NR Jennings, AR>Lomuscio , S Parsons, M. Wooldridg, C. Sierra,
negotiation in E-Commerce, " Grey Systems and Intelligent Services, 2007.              "Automated Negotiation: prospects, method and challenges, " Group Decision
GSIS 2007. IEEE International Conference on, pp. 1528-1532, 2007                       and Negotiation, vol.10 pp.199-215 2001
[28] C.-W. Hang, Y. Wang and M. P. Singh, "An adaptive probabilistic trust             [42] V. Kumar, "Algorithms for constraint-satisfaction problems: a survey, "
model and its evaluation, "Proceedings of the 7th international joint                  AI Mag., vol. 13, pp. 32-44, 1992
conference on Autonomous agents and multiagent systems, vol.3, pp.1485-                [43] D. J. Kim, Y. I. Song, S. B. Braynov and H. R. Rao, "A
1488, 2008                                                                             multidimensional trust formation model in B-to-C e-commerce: a conceptual
[29] L.-k. Soh and J. Luo, "Combining individual and cooperative learning              framework and content analyses of academia/practitioner perspectives, "
for multi-agent negotiations, "Proceedings of the second international joint           Decis. Support Syst., vol. 40, pp. 143-165, 2005
conference on Autonomous agents and multiagent systems, pp.1122-1123                   [44] J. Ahn, X. Sui, D. DeAngelis and K. S. Barber, "Identifying beneficial
,2003                                                                                  teammates using multi-dimensional trust, "Proceedings of the 7th international
[30] S. Sen, A. Gursel and S. Airiau, "Learning to identify beneficial                 joint conference on Autonomous agents and multiagent systems, vol. 3pp.
partners, " AAMAS07 ’07 Honolulu, HI USA, 2007                                         2008
[31] S. Saha, A. Biswas and S. Sen, "Modeling opponet decision in repeated             [45] M. B. Fayek, I. A. Talkhan and K. S. EL-Masry, "GAMA(Genetic
on-shot negotiations, " AAMAS'05, ACM, vol. july,pp. 25-29.2005                        Algorithm driven Multi-Agents) for E-commerce Integrative Negotiation, vol.
[32] G. Yutao, M. Jörg and W. Christof, "Learning User Preferences for                 pp. 1845-1846, 2009
Multi-attribute Negotiation: An Evolutionary Approach, "Multi-Agent                    [46] N. R.Jennings, K. Sycara and M. Wooldridge, "A Roadmap of Agent
Systems and Applications III, pp. 1067-1067, 2003                                      Research and Development, " Autonomous Agents and Multi-Agent Systems,
[33] n. s. choo, machine learning approach for optimizing negotiation agents,          vol. 1, pp. 7-38, 1998
UniversityPutraMalaysia, pp.167, 2007,                                                 [47] V.-W. Soo and C.-A. Hung, "On-Line incremental learning in bilateral
[34] T. D. Nguyen and N. R. Jennings., " A heuristic model for concurrent bi-          multi-Issue negotiation, " AAMAS’02, vol. pp. July 15.2002
lateral negotiations in incomplete information settings, " Proceedings of the          [48] S. Park and S.-B. Yang, "An Efficient Automated Negotiation System
Eighteenth International Joint Conference on Artificial Intelligence 2003, vol.        Using Multi-attributes in the Online Environment, " lecture Notes In
pp. 1467–1469, 2003                                                                    Computer Science, vol. 3140, pp. 544–557, 2004
[35] T. Ebadi, M. Purvis and M. Purvis, "Finding Interaction Partners Using            [49] P. M. C. Samuel, L. Jiming and S.-P. Chan, "Evolutionary Negotiation
Attitude-Based Decision Strategies, "Proceedings of the 2008 Ninth                     in Agent-Mediated Commerce, " lecture Notes In Computer Science, vol.
International Conference on Parallel and Distributed Computing, Applications           2252, pp. 224–234, 2001
and Technologies, vol. pp. 2008
[36] S. Kraus, "Beliefs, time and incomplete information in multiple
encounter negotiations among autonomous agents, " Annals of Mathematics                                           AUTHORS PROFILE
and Artificial Intelligence vol. 20, pp.1-4, 1997
[37] N. Griffiths, "Task delegation using experience-based multi-dimensional
                                                                                                               Sahar Ebadi is currently doing her Master degree in
trust, "Proceedings of the fourth international joint conference on Autonomous                                 Faculty of Computer Science and Information
agents and multiagent systems,vol.11, pp. 489-496, 2005                                                        Technology, UPM. Sahar has received her B.Sc in
                                                                                                               software computer engineering field in 2006 from
[38] N. Gujral, D. DeAngelis, K. K. Fullam and K. S. Barber, "Modeling
                                                                                                               Iran Azad University. Her research interest includes
Multi-Dimensional Trust, " AAAI07_Materials, vol. 6, pp.35-41, 2007                                            Artificial Intelligence and Autonomous Agents and
[39] S. S.Fatima, M. Wooldridge and N. R. Jennings, "Approximate and                                           Data mining.
online Multi-issue negotiation, " AAMAS'07, pp. 947-954, 2007

                                                                                  20                                    http://sites.google.com/site/ijcsis/
                                                                                                                        ISSN 1947-5500

To top