Building Coalitions of Competitors in the Negotiation of Multiparty
e-Contracts through Consultations
Anderson P. Avila-Santos, Jhonatan Hulse, Daniel S. Kaster and Evandro Baccarin
Department of Computing, University of Londrina, Londrina, PR, Brazil
Keywords:
Multiparty e-Contracts, Negotiation Protocol, Auction, Coalition, Consultation, Multiparty Negotiation,
Fairness.
Abstract:
This paper argues that software agents may build two kinds of coalitions in e-negotiation processes. The first is
the typical one in which the parties define roles, rights, guarantees before the negotiation starts. They act as a
team. Either the whole coalition succeeds in the negotiation or fails. In the second one, addressed by this paper,
the coalition members are competitors. They collaborate exchanging information before the negotiation trying
to align their strategies to some degree. Such collaboration only occurs because there is some particularity
(e.g., nearness) that can optimise their business processes if most of the coalition members succeed in the
negotiation. They aim at maximising their chances of success in the negotiation, but act solo. It is important
to note that the main challenge in this scenario lays on the fact that the coalition members are not bind to the
coalition. They may act within the negotiation differently from what they had agreed previously. This gives
rise to the concept of fairness, which is discussed in this paper. The paper also argues that the materialisation
of coalitions within a negotiation protocol fits better in a multiparty negotiation protocol. Thus, it extends the
SPICA Negotiation Protocol with the so-called consultations. The paper presents a study case that shows that
consultations can be benefic to the suppliers, the industry and the consumers.
1 INTRODUCTION
A coalition is an arrangement of two or more par-
ties who cooperate to attain a mutually desired out-
come (Guo and Lim, 2012). It may leverage a given
particularity common to a few negotiators in a way
that is advantageous to other parties besides those ne-
gotiators themselves.
In previous papers, we presented SPICA’s multi-
party contracts and negotiation protocol. If all clauses
were successfully negotiated, a multiparty contract is
signed by the involved parties. This paper extends
the negotiation protocol allowing that a few negotia-
tors, although competitors, organised in a coalition,
exchange information that may help their own deci-
sion making during the negotiation. Such an exten-
sion is made by means of the so-called consultations.
We argue that with minimal change in our negotiation
protocol, we have open a wide window of possibilities
of new patterns of negotiation.
The problem assessed in this paper is how to allow
members of a coalition exchange information and in-
tentions to achieve some kind of benefit within a mul-
tiparty negotiation, restricted that each of them plays
solo within the very negotiation. As a consequence,
only a subset of the coalition members may be suc-
cessful in the negotiation. One alternative would be
the members sign a subsidiary contract among them,
however, this alternative would oblige every member
to honor the subsidiary contract evenif one or more of
them do not win the subsequent negotiation. The ap-
proach we introduce in the paper is to allow members
of a coalition perform a “draft” negotiation, in a way
they may align their strategies, without obliging them.
This gives rise to the discussion about the fairness of
the coalition members.
To illustrate the problem, consider the negotiation
scenario depicted in Figure 1. There is a food com-
pany (FC) that produces granola. Among other in-
gredients, it needs to buy grapes. There are six grape
producers (G
1
. . . G
6
), however, none can individually
fulfil the whole amount of grapes needed by the in-
dustry. Thus, it will buy grapes from half of them.
Due to the geographic specificities, there are 2 groups
of grape producers: farms G
1
, G
2
and G
3
can share
the transportation means (they are on the fringes of
the road R
1
) and the others cannot. Thus, the farms
G
1
to G
3
intend to take advantage of their nearness
618
Avila-Santos A., Hulse J., Kaster D. and Baccarin E..
Building Coalitions of Competitors in the Negotiation of Multiparty e-Contracts through Consultations.
DOI: 10.5220/0005372506180625
In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 618-625
ISBN: 978-989-758-097-0
Copyright
c
2015 SCITEPRESS (Science and Technology Publications, Lda.)
Figure 1: The Granola Company negotiation scenario.
and build an alliance among them to leverage their
competitiveness against the other producers. Never-
theless, as none of them is sure that will win the nego-
tiation with FC, signing a subsidiary contract to guar-
antee the transportation sharing is too risky.
The main contributions of this paper are: (1) it
proposes an extension of our multiparty negotiation
protocol that allows coalition member to exchange
not binding information about their intentions to im-
prove the negotiation process without exposing sensi-
ble information; (2) it shows that such informationex-
changing among the coalition member benefits other
parties besides the members themselves; (3) it shows
that the coalition members do better when they are
committed to the coalition.
This paper is organised as follows. Section 3
overviews a multiparty contract and a multiparty ne-
gotiation protocol we proposed previously. Section 4
extends SPICA with consultations. Section 5 presents
the implementation scenario described above. Sec-
tion 6 runs several experiments using this negotiation
scenario with and without consultation and assesses
the outputs. Section 7 presents related work. Finally,
Section 8 concludes the paper.
2 COALITIONS OF
COMPETITORS: A NEW
APPROACH
Coalitions among players of a givenindustry is a com-
mon practice. They may associate to make the local
market less inhospitable for all of them, to cut costs,
to increase profits, etc. However, such an association
is previously built in a way to protect their individual
interests. Different types of legal instruments, such
as contracts, covenants and treaties, are used to es-
tablish clear boundaries among the members. They
can carefully choose which information will be shared
and provide penalties in case of a misbehaviour.
There is another type of coalition, perhaps less or-
ganised, less structured in which the association do
not only depend on the members’ will, but also on
the events to come. We name this type as coalition of
competitors. For instance, in the scenario presented
in Sec. 1, a few farms are competitors, however none
of them is able to provide alone the amount of grapes
demanded by the food company. They can be more
competitive working together, expecting that all of
them will provide grapes for the food company. How-
ever, such expectancy will be confirmed (or not) by a
subsequent (future) negotiation in which other play-
ers also take part. In this case, only prudential trust
can be employed.
This paper focuses a few interesting questions that
arises in the second type of coalition, such as: in what
extent can a party trust in another coalition member?
Will it behaviour within the actual negotiation as the
coalition has sketched in advance? Is it worth to take
the risk of taking part of such coalition?
These questions give rise to another concept that
we refer as fairness. It expresses the level of confor-
mance of the actions a negotiator takes comparing to
what it promised to the coalition.
3 SPICA MULTIPARTY
CONTRACT AND
NEGOTIATION PROTOCOL
In this section, we present a glimpse of our multiparty
contracts and negotiation protocol. This brief explica-
tion provides to the reader a few concepts she needs
to understand our proposal.
The negotiation process is guided by a contract
template. Negotiators exchange messages that com-
ply with the SPICA negotiation protocol. One of the
negotiators is the so-called Leader, who coordinates
the negotiation process. If there is an agreement, a
contract instance is produced and signed. The negoti-
ation process may be helped by the so-called Notary.
It is a trustworthy third-party that, e.g., receives and
counts votes.
A contract template consists of a set of clauses
with blanks to be filled in. Such blanks are referred
to by the so-called properties and the negotiation pro-
cess aims at assigning values to them. Thus, a con-
tract instance is a contract template with its properties
successfully negotiated. The obligations (or rights)
stated in a clause may bind (or benefit) several part-
ners. Roughly, a clause is composed of a text that
describes the rights and obligations and two lists of
partners. The description is a plain text, but its words
may be prefixed by an ontology name that elucidates
the intended meaning of such word. Property names
may also be embedded in the text. In addition there
BuildingCoalitionsofCompetitorsintheNegotiationofMultipartye-ContractsthroughConsultations
619
are two lists with identifications: the so-called obliged
partners, i.e., the ones that should cooperate to ac-
complished the clauses provisions; the so-called au-
thorized partners those that will share its benefits.
An illustrative example of a clause for the scenario
proposed in Section 1 after it was negotiated is pre-
sented in Fig. 2. In a nutshell, this clause rules that
farms G1, G2 and G4 (obliged parties) must deliver a
given amount (property QTY) of grapes to FC (autho-
rized party) at a given price per ton (property PTON).
The values agreed within the negotiation are assigned
elsewhere in the contract.
text:
The @OBLIGED will deliver to
@AUTHORIZED the amount of #QTY tons of grapes
at the price of #PTON euros each ton.
obliged:
G1,G2,G4
authorized:
FC
Figure 2: A simplified clause.
Basically, a negotiation process runs as follows.
In general, the negotiation of a contract is a part of a
larger process that is controlled by another entity of
our framework, the so-called Coordination Manager
(CM) (Bacarin et al., 2004). Such a manager demands
that a given contract model be negotiated. A new ne-
gotiation instance is created and a leader negotiator is
chosen. This leader decides how the properties will
be negotiated. The leader chooses the most appropri-
ate style to negotiate the properties of a clause. In
our scenario, six farms competed (within an auction)
to deliver the grapes, but only the winners are nomi-
nated in the obliged list and will cooperateto provided
that total amount of grapes expected by FC, e.g., each
one will provide a third of QTY. Note also, that a few
of them are coalition members (G1 and G2), but G4
is not. A detailed discussion about our proposal for
multiparty contracts is outside the scope of this paper.
The protocol defines the messages and the data ex-
changed among the players of a negotiation. They
convey several parameters that tune up a specific ne-
gotiation, identify the sender and the receivers, and
help establishing correlation among messages. Only
the relevant parameters for the purpose of this paper
are presented.
3.1 Exchanging Data Types
Most of the negotiation messages build on two ba-
sic data types: Request For Proposals (RFPs) and Of-
fers. An RFP invitesanother party to negotiate a set of
properties. A negotiator A sends an RFP to a negotia-
tor B asking for a value for one or more properties.
More specifically, an RFP conveys three pieces of
data: a set of asked properties, a set of assigned prop-
erties and a restriction. The example below shows an
RFP written in a simplified notation. The RFP pro-
poses a value for
QTY
(3), asks a value for property
PTON
, but imposes that it would be lesser than 700.
≪{
QTY:3
}
,
{
PTON
}
,‘PTON
<
700’
≫
A negotiator A proposes a value to one or more
properties sending an Offer to a negotiator B. If ne-
gotiator B accepts it, both negotiators are committed
to the proposed values. A negotiator answers an RFP
sending back an Offer that assigns values to the asked
properties.
The example below shows an Offer that answers
the previous RFP. Note that this Offer does not change
the proposed value for property
QTY
(it should not)
and assigns the value 690 to property
PTON
.
[QTY=3, PTON=690]
RFPs and Offers are used to build several styles
of negotiation that boil down to three basic ones: bar-
gain, ballots, and auctions. Bargains are used within
bilateral negotiation, auctions are used when there
is competition among a few negotiator, and ballots,
when consensus among negotiators are needed. Other
styles (e.g., different flavours of auctions) are ob-
tained from different setups of these basic ones.
3.2 Negotiation Interfaces
The negotiation messages are defined by means of a
set of interfaces (Java, in the current implementation)
with methods to be implemented by the different ne-
gotiation players.
Figure 3 exemplifies the use of some of those in-
terfaces. It depicts the auction messages and the play-
ers. The players are: the Notary, the Leader negotiator
(FC) and other negotiators (farms). The exchanged
messages are represented by arrows (e.g., the first ar-
row depicts the sending of a message from Leader
to Notary). Each arrow has a caption that displays
the message sent (firstAnswerToProposal, in the ex-
ample). This caption presents the name and the main
parameters of the invoked method with a prefix that
identifies its interface (LeaderIF, in the example).
An auction is a sequence of the so-called auction
steps: (1) The Leader sends a firstAnswerToProposal
message asking the Notary to advertise and conduct
an auction. This message inform the Notary how
many bids (nbids) to collect (e.g., 3) within a given in-
terval t (e.g., 10s). The auction’s subject is described
either by an RFP or by an Offer. In the former, the
bidders propose values for asked properties (as usual
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
620
!
Figure 3: The steps of an auction in SPICA.
in English auctions); in the latter, the interested bid-
ders may agree to the proposal (as usual in Dutch
auctions). From now on, we will use RFP auctions.
(2) The Notary accepts the task (message willAuc-
tion) and broadcasts the RFP (message requestPro-
posal) and waits as demanded. The negotiators re-
ceive the RFP and (3) send Offers to the Notary in
response (message receiveProposal). The Notary col-
lects them and (4) sends them to the Leader (message
collectedAnswersToProposal). If the Leader guesses
it can have better bids, it may run another auction step
enhancing the restrictions stated in the RFP: in the ex-
ample it would try to decrease the grape price. For in-
stance, if the cheapest bid received within the auction
step proposed PTON=690, FC would ask the Notary
to conduct another step imposing that PTON<690.
Ballots have a similar pattern. Briefly, votes are
sent instead of bids. Bargains happen between only
two negotiators.
4 INCLUDING CONSULTATIONS
IN e-NEGOTIATION PATTERNS
The negotiators in a coalition use the so-called consul-
tations to exchange information about their intentions
or trying to establishing mutual consensus during the
negotiation process. It is important to note that the
consultation process happens “out the boundaries” of
the negotiation process and its results are not binding.
It means that a negotiator’s actions within the negoti-
ation process need not comply to its response within
a previous consultation.
We augmented the protocol with the so-called
consultation messages, specified by means of a few
new interfaces, which uses two exchanging data
types: Requests For Information (RFIs) and Informa-
tion (Info). We also provided a trusted third party to
help the consultation interactions, the so-called Medi-
ator.
An RFI is very similar to an RFP: it asks values
for properties, but also lower and upper bounds for
them. An Info is similar to an Offer: it proposes val-
ues for asked properties and also informs upper and
Figure 4: Pattern Summary of Intentions in SPICA.
lower bounds for them, however, the negotiator which
issued an Info is not committed to it.
The consultations may happen in two different
ways: (a) aiming at convergence of intentions; or (b)
agreeing on proposal. These patterns are described in
the following.
In convergence of intentions, the negotiators share
information and each of them makes its better effort
trying to align their strategies. For instance, in our
negotiation scenario, the negotiators inform their in-
dividual transportation cost and based the aggregated
mean value, they estimate how much they would save
if they shared the transportation means. Thus, each
negotiator make a more attractive individual offer to
the food company. This pattern as two flavours: (a)
Summary of Intentions; and (b) Burst of Intentions.
• Summary of Intentions: Figure 4 depicts this
kind of consultation. It is started by means of
the message consultTendency. Such message is
answered by a subsequent receiveVeiledIntention
message. This message does not disclose the
negotiator’s intentions to the group, but sends
them to the Mediator who summarises all received
intentions (e.g., calculate the mean value) and
broadcasts the summary to the group (message re-
ceiveIntentionSummary).
• Burst of Intentions: A negotiator asks other ne-
gotiators about their intentions regarding a spe-
cific issue. All negotiators broadcast their inten-
tions. It is not helped by a Mediator.
In the Agreeing on Proposal pattern, the negotia-
tors aim at a consensual answer for a specific issue. It
is similar to the Summary of Intentions pattern, but a
ballot is run to find such consensus.
5 TEST CASE: GRANOLA
COMPANY
This section presents the implementation of the ne-
gotiation scenario in Sec. 1. For brevity’s sake, the
focus of the paper lays on the negotiation of one item
(grapes) and the consultation pattern is the Summary
of Intentions.
BuildingCoalitionsofCompetitorsintheNegotiationofMultipartye-ContractsthroughConsultations
621
The negotiation Leader is the food company (FC)
which provides the contract model. FC wants to pay
the least price. Thus, it runs several auction steps
asking decreasing prices for the grapes. At the end,
FC will agree with the three best bids to provide the
grapes. Note that, there can be zero to three CC mem-
bers among the providers.
By their turn, the grape producers use decision
tables to assess the proposals they receive: for each
property each negotiator has a value (or a range of
values) it considers a “good deal” (i.e., the expected
value – EV) and for which it will always agree upon.
There are also values the negotiator will never agree
upon. Values in-between will be accepted with spe-
cific probabilities: the nearer to the expected value,
the higher the acceptance probability. Decision tables
keep such values and probabilities. Such strategy, al-
though simple, is quite effective to simulate the nego-
tiation.
A few of the grape producers can share the trans-
portation to enhance competitiveness (G
1
, G
2
and
G
3
). Thus, they compose a consultation community
(CC). The community’s negotiators may change (tem-
porarily) their expected values (EV) according to the
consultation result. It was implemented as follows.
Each CC member, besides its decision table, keeps
two pieces of information: the total production cost
(ToPC) and the percentage the transportation con-
tributes to the total cost (TRCp). Thus, the nego-
tiator’s expected profit (EP) is easily calculated as a
function of TPC, TRCp and EV.
Before negotiating the grapes, the CC runs a con-
sultation of the kind summary of intentions, as fol-
lows. (i) Each negotiator sends to the Mediator its
transportation cost. (ii) The Mediator returns to the
CC members the mean value (TRCm). (iii) The ne-
gotiators assume that TRCm will be the total trans-
portation cost for them. Such assumption considers
that a single truck load is able to transport the grapes
of all those producers, therefore it would be under-
used by a single producer. Since they will share the
transportation, each negotiator will chip in a third of
this value (they are 3 members). (iv) Next, they re-
calculate their expected value to have the same profit
margin they would have previously (EV’). (v) Finally,
the CC members partake the negotiation using the ad-
justed decision table.
It is noteworthy that the presentedconsultation did
not considered a property of the contract. Instead
of consulting about the value of the grapes (contract
property), the consultation was about another param-
eter that the negotiators could minimise if they could
take advantage of their geographical nearness (i.e, the
transportation cost).
The reasons for using a Summary of Intentions
were twofold: it aimed at maximising the win-win ap-
proach among the CC members and also hiding sen-
sible values.
Once consultation is not binding, the commitment
a negotiator has regarding to CC may vary. We refer
to the level of commitment of a negotiator as its fair-
ness: the more committed, the fairer. The measure of
fairness is a real number in the range [0, 1]: 0 means
that the negotiator is not committed at all to the con-
sultation results; 1 means that the negotiator is fully
committed to such results. To evaluate how the fair-
ness behaviour affects the performance of a CC, each
CC member keeps another piece of information: its
fairness (F). Every member will change its expected
value to ((F ∗ EV
′
) + (1− F) ∗ EV). Thus, in one ex-
treme a fully fair CC member (F = 1) will use the
recalculated expected value (EV’) as described previ-
ously. Conversely, a fully unfair member (F = 0) will
not change its decision table.
6 ASSESSMENT
We ran several experiments to assess the outcomes of
using consultations in the test case.
The experiments presented in this section address
the following questions.
1. Have the coalition members provided grapes to
the industry as a group more often than individ-
ually?
2. Has the profit of the each coalition member in-
creased when compared to the profit the farm had
before becoming a coalition member?
3. Has the procurement price decreased (for the in-
dustry)?
4. In what degree the level of commitment (fair-
ness) of the coalition members influence the re-
sults for a given individual member and for the
whole coalition?
The first and second questions assess whether the
coalition was advantageous for its members. The
third question tries to determine if the coalition con-
ferred benefits to the consumers. It supposes that,
once the industry obtains cheaper grapes, its products
can also cost less to the consumers. The last question
assesses the level of risk of using consultations, since
they do not imply strict commitments.
We built a basic setup that aimed at assessing the
first three questions by means of a batch of execu-
tions. In this setup, all CC members had their fairness
set to 1 (fully committed). To evaluate fairness, the
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
622
basic setup was run a second time now varying the
degree of fairness of the coalition members. All ex-
periments were executed 20 times (half with consul-
tations, half without them). In the following, we plot
and comment the achieved results.
With regard to Question 1, Figure 5 shows how
many auctions a negotiator won (in percentage) com-
paring experiments with and without consultations.
That figure shows that the CC members, when stuck
to the consultation (F = 1), did no worse than not us-
ing consultation. In fact, all did better, but G3 that
had the same outcome. This result to G
3
is because it
already was quite competitive. This figure also shows
that G
6
is the most competitive negotiator among the
non CC members.
Figure 5: Winning negotiators with and without consulta-
tions.
Figure 6: Total profit achieved with and without consulta-
tions.
Question 2 is addressed in Figure 6, which shows
the sum of the profits attained by each negotiator
within 10 experiments. It is easy to notice that the
non CC members did much worse when the other ne-
gotiators were running consultations. It is notewor-
thy that G
3
did a slight bit worse when partaking the
consultation (again, because it was already competi-
tive). However, it would not know this fact before-
hand. Moreover, if G
3
is compared to G
6
(both the
most competitive in their groups), we can see that G
6
did much worse. Thus, if G
3
would have refused to
participate in the CC, G
6
could do it somehow and
their outcomes would be inverted.
Regarding Question 3, Figure 7 sums up the to-
tal value paid by FC within 10 experiments. FC has
clearly paid less when consultation was active. We
also wonder if the total value paid by FC was corre-
lated to the number of CC members that won a given
auction. Figure 8 was used to investigate such hy-
pothesis. This figure comprises bars and lines. A bar
represents to total value paid by FC in a given auc-
tion, and the lines shows how many CC members won
such an auction. For instance, in the first auction with
consultation FC paid a bit more than 60,000 to the
winners and all three CC members won that auction.
Conversely, in the first auction without consultations,
FC paid at most 60,000 to the winners and only one
of them was a CC member. There is no clear correla-
tion between the cost and the number of winning CC
members. However, this figure confirms that the only
existence of consultations brings the cost down due to
the competitiveness rising.
Figure 7: Total cost for the FC (∗10
3
).
1 2 3 4 5 6 7 8 9 10
45
50
55
60
65
70
Auction
Cost (x10^3)
0
1
2
3
Participants
Without consultation
With consultation
Without consultation
With consultation
Figure 8: Cost by auction instance and number of winning
coalition members with and without consultations.
In order to assess how fairness influenced the ne-
gotiation process (Question 4), we ran two experi-
ments. In the first (Figure 9), all CC members were
set with the same fairness and the total profit of
the CC member (after 10 negotiations) was summed
(solid blue line). Similarly, the profit of the non-
members was summed (dashed green line). This was
repeated for different fairness (0, 0.1, 0.2, ..., 1). For
BuildingCoalitionsofCompetitorsintheNegotiationofMultipartye-ContractsthroughConsultations
623
instance, when the CC members were set with F =
0.6, the total profit of the CC members was higher
than 80,000 and for non-member, less than 20,000.
This figure shows that the CC members do better al-
together when they are fair. In the second experiment
(Figure 10), two CC members had the same fairness
(F = 0.5) and the remainder (G
1
) had its fairness var-
ied. For each level of fairness, we ran 10 negotiations
and calculated the total profit G
1
gained. Its clear the
ascending pattern as G
1
become more fair, reaching
the maximum profit of more than 30,000 at F = 0.8.
! " #
"
$%
&
'
Figure 9: Total profit varying the fairness of all members.
Figure 10: Profit varying the fairness of only one member.
7 RELATED WORK
The notion of coalitions has been studied by the
game theory community for a long time and it has
proved to be a useful strategy (Horling and Lesser,
2004). It has gaining increasing interest by the Infor-
mation Technology community as an interaction pat-
tern to develop complex multi-agent systems. There
are a few software methodologies tailored to agent-
oriented systems that allow implementing this or-
ganisational paradigm (Isern et al., 2011). How-
ever, several proposals regarding automatic negotia-
tion employs case-specific implementations, such as
(Yu et al., 2013). Works that present more structured
implementations usually rely on negotiation protocols
and frameworks.
Most of the negotiation protocols are based on
bilateral interactions, e.g., (Shakun, 2005) and (Br-
zostowski and Wachowicz, 2014). However, collab-
orative organisations need negotiation protocols that
are multiparty and interactive (Darko-Ampem et al.,
2006). There are a few multiparty negotiation pro-
tocols in literature, e.g., (Fujita et al., 2012), (Klenk
et al., 2012) and (Szapiro and Szufel, 2014). These
works differ from ours, since they do not integrate
several styles of negotiation, namely,bargain, ballots
and auctions for the negotiation of a given multiparty
contract. In addition, our negotiation protocol also
provides a non binding coalition mechanism.
According to (Peleteiro, 2014), coalition forma-
tion is a process in which agents associate to achieve
a goal or to increase their performance, and is guided
by rules of formation. She also mentions the problem
of stability of a coalition, i.e., the level of incentive a
member has to withdraw (internal stability) the coali-
tion and a non-member has to join it (external stabil-
ity). Another aspect she considers is if the coalition is
statically or dynamically formed. In contrast with the
former, in the latter the agents are constantly willing
to change the coalition they belong to. A coalition
may be ruled by a leader that imposes its strategy.
According to (Yu et al., 2013), another issue regard-
ing coalition formation is how to allocate the profit
among its members.
Our protocol does not tackle the coalition forma-
tion problem, but supposes that it was somehow built
prior the negotiation and, most important, it does not
bind the members of the coalition.
Stability of coalitions is an issue worth of discus-
sion. In general, a coalition of agents behaves as a
block. Thus, internal and external stability is a clear
concept. Our proposal is quite different. A coali-
tion of competitors has three distinct moments. In
the first moment, the coalition members exchange in-
formation in order to be somehow aligned during the
actual negotiation. The second is the very negotia-
tion. In this phase, the coalition members may be
fair to that alignment or not. If they are fully unfair,
we can consider that they left the coalition (no inter-
nal stability). Conversely, if they are fully fair, they
stayed in the coalition. However, as we mentioned
previously, a coalition member can be partially fair.
In this case, the concept of stability becomes blurred.
Finally, the third moment is the coalition realisation,
i.e., the members that succeed the negotiation become
the signatories of the produced contract. Note that,
this realisation may also be partial, once a few of the
coalition members may fail in the negotiation.
Another difference of our approach for coalitions
is that there is no leader ruling or coordinating the
acts of the coalition members and there is no direct
allocation of profits among the members. In our case,
profits may be earned indirectly, e.g., by increasing
competitiveness. In addition, (Guo and Lim, 2012)
argues that coalitions are just formed to reduce the
ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems
624
intrinsic complexityof multiparty negotiations. In our
proposal, it is not an issue.
8 CONCLUSION
A coalition is referred as in the literature as an or-
ganised, framed, protected, as a solid and monolithic
block. It is materialised before the negotiation takes
place. It is as if the coalition was a big negotiator act-
ing in behalf of its individual parties.
This paper focused on a second kind of coalition,
that we called coalitions of competitors, which is a
more challenging, risky and fluid association of nego-
tiators. It differs from the previous one in four basic
features. First, the coalition is not formalised and ma-
terialised beforehand. In fact, a few negotiators intend
to build a coalition. They exchange information either
before or during the negotiation process to maximise
their chances of success. However, had they tuned
up their decision processes, each one negotiates by it-
self. Thus, the coalition is actually realised after the
negotiation. This leads to a second difference: it can
produce partial coalitions, as just a few of the coali-
tion members end up succeeding in the negotiation.
The third is that, since the coalition may succeed only
partially, a negotiator does not wish to commit to an
agreement that will bind it independently of the result
of the negotiation. Finally, as consequence, any coali-
tion member may act within the actual negotiation dif-
ferently from the previous coalition agreement.
We reified coalitions through the SPICA Negoti-
ation Protocol. This protocol was suitable for this
purpose once it implements multiparty negotiation
of multiparty contracts. The protocol was extended
to allow coalition members to exchange information
within a negotiation process by means of consulta-
tions. The effectiveness of such extension was as-
sessed by means of an experimental negotiation sce-
nario. The results were twofold: (a) the execution of
a bunch of experiments showed that exchanging in-
formation different from those present in the multi-
party contract being negotiated improved the outcome
to the negotiators (CC members and FC) as well as the
consumers of the product delivered by FC; (b) in gen-
eral, do better the members that are fair to the coali-
tion.
Future work includes assessing the other two con-
sultation patterns that were just mentioned in this pa-
per and improving the intelligence of the negotiators.
ACKNOWLEDGEMENTS
This research was partially supported by agencies
Fundac¸˜ao Arauc´aria, Capes and CNPq. We also thank
Prof. Caetano Traina Jr and Prof. Agma J.M. Traina
(GBDI, ICMC/USP) for sharing their resources.
REFERENCES
Bacarin, E., Medeiros, C., and Madeira, E. (2004). A Col-
laborative Model for Agricultural Supply Chains. In
CoopIS 2004, LNCS 3290, pages 319–336.
Brzostowski, J. and Wachowicz, T. (2014). Negomanage:
A system for supporting bilateral negotiations. Group
Decision and Negotiation, 23(3):463–496.
Darko-Ampem, S., Katsoufi, M., and Giambiagi, P. (2006).
Secure negotiation in virtual organizations. In
EDOCW ’06, pages 48–55. IEEE.
Fujita, K., Ito, T., and Klein, M. (2012). A secure and fair
protocol that addresses weaknesses of the nash bar-
gaining solution in nonlinear negotiation. Group De-
cis. and Negot., 21(1):29–47.
Guo, X. and Lim, J. (2012). Decision support for online
group negotiation: Design, implementation, and effi-
cacy. Decision Support Systems, 54(1):362 – 371.
Horling, B. and Lesser, V. R. (2004). A survey of multi-
agent organizational paradigms. Knowledge Eng. Re-
view, 19(4):281–316.
Isern, D., S´anchez, D., and Moreno, A. (2011). Organiza-
tional structures supported by agent-oriented method-
ologies. J. of Sys. and Soft., 84(2):169–184.
Klenk, A., Beck-Greinwald, A., Angst, H., and Carle,
G. (2012). Iterative multi-party agreement negotia-
tion for establishing collaborations. Service Oriented
Computing and Applications, 6(4):321–335.
Peleteiro, A. (2014). Dynamic Coalition Formation Mech-
anisms for Enacting and Sustaining Cooperation in
Multi-agent Systems (MAS). PhD thesis, Univ. of
Vigo.
Shakun, M. (2005). Multi-bilateral multi-issue e-nego-
tiation in e-commerce with a tit-for-tat computer
agent. Group Decis. and Negot., 14(5):383–392.
Szapiro, T. and Szufel, P. (2014). Simulated negotiation
outcomes through recommendation crowding. Group
Decision and Negotiation, 23(3):443–461.
Yu, F., Kaihara, T., and Fujii, N. (2013). Coalition for-
mation based multi-item multi-attribute negotiation of
supply chain networks. Procedia {CIRP}, 7:85–90.
BuildingCoalitionsofCompetitorsintheNegotiationofMultipartye-ContractsthroughConsultations
625
