A Negotiation Optimization Strategy of Collaborative

Research ArticleA Negotiation Optimization Strategy of CollaborativeProcurement with Supply Chain Based on Multi-Agent System

Chouyong Chen and Chao Xu

Management School Hangzhou Dianzi University Hangzhou Zhejiang 310018 China

Correspondence should be addressed to Chao Xu chaoxu hdu163com

Received 13 March 2018 Revised 29 July 2018 Accepted 5 August 2018 Published 26 August 2018

Academic Editor Luciano Caroprese

Copyright copy 2018 ChouyongChen andChaoXuThis is an open access article distributed under theCreativeCommonsAttributionLicensewhichpermits unrestricteduse distribution and reproduction in anymedium provided the original work is properly cited

In the process of collaborative procurement buyers and suppliers are prone to conflict in cooperation due to differences in needs andpreferences Negotiation is a crucial way to resolve the conflictAimed at ameliorating the situations of underdeveloped self-adaptivelearning effect of current collaborative procurement negotiation this paper constructs a negotiation model based on multi-agentsystem and proposes a negotiation optimization strategy combined with machine learning It provides a novel perspective for theanalysis of intelligent SCM The experimental results suggest that the proposed strategy improves the success rate of self-adaptivelearning and joint utility of agents compared with the strategy of single learning machine and it achieves win-win cooperationbetween purchasing enterprise and supplier

1 Instruction

Information technology has enabled and in some casesforced enterprises to reorient their internal capabilities andto redefine their business models to develop e-commercetechniques In order to attain timely responsiveness and toproffer higher service level constructive cooperation amongpartners in supply chain is critical in any endeavor to ame-liorate disruptions and mitigate risks [1] A small number ofsuccessful contemporary associations have transformed froman opportunistic doctrine of cooperation to a synergisticethos and integrated their supply chain procedures

The synergism of Cluster Supply Chain (CSC) compris-ing collaborative manufacture collaborative procurementcollaborative logistics and collaborative inventory is the cou-pling organizing form between industrial cluster and supplychain And it helps small andmedium-sized enterprise (SME)shorten the transaction cycles and reduce costs Procurementdirectly affecting the production and operation is the keylink in the development of the whole enterprise In thefierce market competition therefore the purchasing modegradually shifts from traditional independent purchasingthat faces the problems of small quantity discount lowbargaining power and slow response to customer demand tocollaborative procurement

In this article we consider a distributed supply chain (SC)in that each member seeks to optimize personal performanceand independently plans his business A large measure ofsupply chain managements (SCM) have to communicate andnegotiate effectively with SC members In the process ofcollaborative purchasing buyers and suppliers are prone toconflict in cooperation due to differences in needs and prefer-ences Negotiation is a crucial way to resolve the conflict andan effective mechanism for supply chain coordination andcooperation It has been demonstrated that the informationsharing between the buyers and suppliers ensures effectivesupplier participation and enhances mutual understandingwhich contributes to more excellent performance over therivals [2] Negotiation is reckoned to be a sound approachfor participators to exchange messages understand otherperspectives and identify new order alternatives based onthe information and knowledge learned in the process Andit allows enterprises in the CSC to prevent both self-interestand local optimization of finicky partner to proceed to theoptimization of objectives of all participators and to achievea win-win situation in SCM

Former researchers have paid much attention to negotia-tion problem over the past decade and proposed some salientmodelsThemajority of models primarily use either methodsof the improved algorithms or game-theoretic techniques

HindawiMathematical Problems in EngineeringVolume 2018 Article ID 4653648 8 pageshttpsdoiorg10115520184653648

2 Mathematical Problems in Engineering

as a basis to formulate autonomous negotiation Howeverthose approaches are considered to be complicated to spreadto widespread problem fields due to the uncertainty andcomplexity in real-world negotiation This paper amelioratesthe negotiation model combining multi-agent system (MAS)with machine learning for further tackling the conflict incooperation New model provides a buyer with a method forpurchasing a product systematically And it helps in achievinga win-win cooperation between two sides during the processof collaborative procurement with supply chain as far asfeasible

The remainder of the paper is structured as follows Sec-tion 2 shows literature review Section 3 recalls some generalconcepts of key techniques Section 4 is devoted to a nego-tiation model of collaborative procurement based on MASSection 5 describes a self-adaptive negotiation optimizationstrategy combined with dynamic selective ensemble learningFinally the experiments design results and concludingremarks are presented in Sections 6 and 7 respectively

2 Literature Review

Given that research on negotiation of collaborative pro-curement is new and largely fragmented it is practicallyparamount to arouse individualsrsquo attention Previous studieshave nevertheless proposed a basic model of supply chainsand a negotiation strategy for solving conflicts in consider-ation of efficiency and cost A multi-objective cooperativeproductionndashdistribution planning model was formulated byJolai et al [3] applying the fuzzy goal programming approachto maximize the gains of all participators To discover theoptimal solutions of resource allocation Lin et al [4] rec-ommended a collaborative negotiation mechanism that wasbuilt on price schedules decomposition algorithm But thepopular methods to research the negotiationrsquos conundrumin SCM involve Game Theory and artificial intelligence(AI) Game theorists deem the negotiation as an incompletedynamic information game and attempt to settle the gameby offering some predictions on certain conditions [5]Primary methodological tools of Game Theory are Nashgame [6] and Stackelberg game [7] which concentrate onthe sequential and simultaneous decision-making of multipleplayers respectively For those relevant analytic modelingstudies the problem is analyzed mostly from a theoreticalperspective Despite being extremely successful in a quantityof situations the game theoretical approach is considered tobe difficult to spread to universal problem fields owing to theuncertainty and complexity in real-world negotiation

Compared to Game Theory participants that bargainwith consideration of human preference and thoughts couldbe considerably represented by the agent technology whichis a branch of AI [8] The use of information and communi-cation technology tools offering the capacities of customersensitivity information sharing and process integration isobserved as the uppermost enabler for this collaborative per-ceptionrsquos realization [9] In computer science an agent is gen-erally considered as a software entity which is autonomous tocommunicate and coordinatewith other agents to accomplishits design objectives Consequently multi-agent simulation

modeling which originated from AI is suitable for theconduction of distributed system and has certain advantagesin being testable quantifiable and efficient It is superior inexpansibility is easy to configure and has been widely usedin the SCM Kwon et al [10] constructed an integrated frame-work that was based on multi-agent cooperation and case-based reasoning to help address emerging uncertainties Linet al [11] demonstrated a supply chain coordination modelof multi-agent and put forward a conflict solution methodbuilt on constraint satisfaction algorithm due to the differentformof demand Considering the conflict between businessescaused by the difference of information asymmetry and goalsBehdani et al [12] developed a negotiation method based onmulti-agent in the condition that demand is uncertain Thesignificance of addressing negotiation mechanisms for col-laborative matters is shown by the discussed literatures Thecombination of negotiation model and optimization technol-ogy is requisite to help negotiators achieve optimal selections

In order to better promote the agentrsquos self-adaptive nego-tiation ability an army of scholars have begun to introducemachine learning into the negotiation Bayesian Learningestimates the probability distribution of opponent negotia-tion parameters and preferences and adaptively adjusts theconcession strategy [13] Q-Learning generates the optimalnegotiation strategy by calculating the utility cumulativevalue [14] Radial Basis Function (RBF) neural network iscapable of optimizing the Actor-Critic learning algorithm topredict and amend the concession magnitude of agents [15]Unfortunately previous self-adaptive negotiation is built on asingle or integrated learning machine to draw the final result[16] Selective ensemble learning improves the efficiency ofgeneral integrated learning machine by eliminating the lessaccurate ones in sublearning models [17]

This paper is built on our previous work in the fieldof automated negotiation In particular it lays the foun-dation for accomplishing an experiment to investigate theperformance of agent which is operating in the supply chainsystem and equipped with our negotiation model The maincontribution consists of constructing a negotiation modelconcerning collaborative procurement based on MAS byanalyzing the characteristics of multilateral transact andproposing a negotiation strategy founded on dynamic selec-tive ensemble learning We exploited supply chain analysisdetailedly that was based on agent technology which detectsnovel patterns through the improved data mining techniquesand provides a new perspective for the analysis of intelligentSCMMoreover agent jobwas led by this association betweenintelligent agents and machine learning to do faster andbetter And the negotiation strategy has also potential for bigdata decrement and compression

3 Methods

31 Machine Learning Machine learning gradually becomesan irreplaceable method for processing data in the big dataera As an embranchment of AI it has entered foreland ofthe mainstream computer sciencersquos research that often usesstatistical techniques to give agents the ability to learn withdata without being explicitly programmedMachine learning

Mathematical Problems in Engineering 3

has substantial connections with mathematical optimizationwhich delivers theory application domains and methodsto the field Moreover it is a popular method practiced todevise complicated models and algorithms for predictionThese analytical models permit researchers to find results andauthentic decisions and reveal hidden insights via learningfrom historical relationships and tendencies in the data

32 K-Means Clustering K-means clustering an unsuper-vised learning is fundamentally a partitioning method thatis utilized to analyze data and treat the datarsquos observations asobjects on the basis of locations and distance between diverseinput data points It helps to partition the undisposed objectsinto mutually exclusive clusters (K) so that objects remain asclose as possible to each other within individual cluster but asfar as possible from other clustersrsquo objects

33 Support Vector Machine Support Vector Machine(SVM) introduced by Vapnik is originated from the theoryof structural risk minimization belonging to statisticallearning theory The essential idea of SVM is to map inputvectors into a high dimensional feature space and constructthe optimal separating hyperplane in this space SVM tries tominimize an upper bound of the generalization error bymax-imizing the margin between the test data and the separatinghyperplane [18] It has severalmerits (1) Aunique hyperplanemaximizing the margin of separation between the classes canbe discovered by SVM so it has a good ability of robustness(2) SVMrsquos power is to use kernel function to transform datafrom the low dimension space to the high dimension spaceand create a linear binary classifier (3) The solving of SVMis a convex programming problem and its local optimumis selected as the global optimum In the field of machinelearning models combined with learning algorithms foranalyzing and classifying data are represented by SVM

4 Negotiation Model of CollaborativeProcurement Based on MAS

One of the most distinguishing advantages of using MAS forSCM is the dynamic supply chain construction via automatednegotiation between agents In the MAS the coordinatoragent is introduced to regulate multiple buyer and selleragents A distributed negotiation model based on MAS isdemonstrated in Figure 1 The model assists enterprises inchoosing the most suitable suppliers quickly efficiently andeconomically The system consists of 3 mutually coordinatedagents CA represents the supplier agent PA the purchasingenterprise agent of industrial cluster and MA the brokeragent of collaborative purchasing service Agents participat-ing in the negotiation must register with MA (such as an e-commerce platform) in advance and configure a unique IDTheMAmanages various information in the negotiation pro-cess and coordinates the communication between the agentsThe selection of the supplier is done with the assistance oftheMA and repeated negotiation between the PA and the CA(the types of messages used by the agents in the negotiationprocess are shown in Table 1)

MA (i) It promptly registers verifies and updatesinformation about registered agents (ii) It duly publishesforwards and organizes messages (iii) It comprehensivelyutilize real-time environment and enterprise data to evaluatethe operation of businesses

PA If Reply is received PA will compare the propertyvalues of the products given by the participating CA withaccredited ones and then send Improve to the nonoptimalCA Subsequently it selects CA whose values are no lessthan the threshold as a candidate supplier If there is noqualified supplier purchasing enterprise will modify therelevant threshold and renegotiate with all suppliers Finallythe result opted for is sent to the MA with Selection Afterreceiving Confirm if the CA is found to have objected tothe negotiation result check the modification and resendImprove until no objection occurs

CA After monitoring Announce published by the MA ifthe requirements of order are met deliver the Bid to partici-pate in the negotiation In the event of corresponding valuessuggested by the PA being acceptable during Adjust CAsends a newBid or else emits Reject Eventually when receiv-ing Result the selected CA checks the content of the protocoland if there is no objection the Accept is fed back Otherwisethe Refuse is transmitted to point out the problem

The specific negotiation process is showed in Figure 2

5 Self-Adaptive NegotiationOptimization Strategy

51 Negotiation Parameter Negotiation parameters consistedof four elements which are proposed and explained in Table 2

NM = A PwU (1)

52 Concessional Learning Based on Dynamic SelectiveEnsemble of SVM According to the current negotiationissues the nearest neighbor sample set is used as the trainingsample to evaluate the performance of each submodel andselect the better ones In the negotiation K-means algorithmis adopted for each negotiation issue and the k samplesubsets are found as the training datasets And the SupportVector Machine (SVM) is used to learn the concessionamplitude in each evaluation sample Taking root-mean-square error (RMSE) as the evaluation criterion we eliminatesome submodels with poor performance The combinationweight is calculated and the final dynamic selective SVMmodel is established

(1) K-means algorithm generates evaluation datasets 119875119902is negotiation sequence to be predicted and its number of thenearest neighbor sample in the data set 119875119871 is k and the first ksamples 119875k can be got by calculating the Euclidean Distance119875119863 between 119875119902 and the sample points 119875i

119875119863 (119875119902 119875119894) = radicsum119894isin119871

(119875119902 minus 119875119894)2 (2)

(2) Input sample set 119875k and estimate concession amplitudewith SVM Assume that negotiation values of 119860119862 and 119860119875 inround t and issue j are denoted as 119875119862119905 and 119875119875119905 respectively


Figure 1 MAS negotiation model of collaborative procurement

Table 1 Instructions of related messages

Agent Message Description of MessageAnnounce Publish PArsquos requirement information to registered AgentsAdjust Forward PArsquos improved requests to CA

MA Result Inform of results and send the agreed protocol contents to CAReply Send the product and enterprise information to PA

Confirm Transform the confirmation messages or protocol modification informationRequest Ask MA to release massage to corresponding CAInquire Consult MA for information about the supplier

PA Improve Request PA to improve the relevant attributes on MASelection Post results and agreements to selected PAReject Notify the MA not to participate in the consultation

CA Bid Give information of the product and the enterprise or submit the improved relevantattributes

Accept Acquaint MA the accepted agreementsRefuse Object the protocol or request for verification

Table 2 Instructions of negotiation parameters

Parameters InstructionsA 119860119862 represents Supplier 119860P Industrial cluster buyerP Issue value of negotiationw Weight vector of the issueU Utility value of the issue

and Δ119863119905 is the negotiation difference between 119860119862 and 119860119875obtained by (3)The average concession amplitudes of119860119862 119860119875for the first t rounds are 119862119862119905 119862119875119905 As inputs to the SVM t Δ119863119905119862119862119905 and 119862119875119905 are mapped to the high dimensional space usingthe Radial Basis Function119867119905 = (120593(119905) 120593(Δ119863119905) 120593(119862119862119905 ) 120593(119862119875119905 ))119862119875119905+1 is the output variable of the linear regression functionobtained by (5)

Δ119863119905 = 10038161003816100381610038161003816119875119862119905 minus 119875119875119905 10038161003816100381610038161003816 (3)

119862119875119905 = 119905sum119894=2

Δ119863119894119875119875119894minus1 (4)

119862119875119905+1 = 119908119879 lowast (120593 (119905) 120593 (Δ119863119905) 120593 (119862119875119905 ) 120593 (119862119862119905 )) + 119887 (5)

where 119908119879 is the weight vector of 4 input variables and 119887 is aoffset value

The error 120576 between predicted value y and function value119862119875119905+1could be calculated by (6) If the error 120576 is regarded asan error-free fitting then we can get the nonlinear regressionfunction as (7) of the concession amplitude 119862119875119905+1of theopponent in round t+1 After the equivalent substitution wecan get the final regression function as (8)

max 0 10038161003816100381610038161003816119910 minus 119862119875119905+110038161003816100381610038161003816 minus 120576 (6)

119862119875t+1 = 119899sum119895=1

(119886119895 minus 1198861015840j )119870 (119867119905 119867119905minus1) + 119887 (7)

119862119875t+1 = 119899sum119895=1

119886119895 expminus1003817100381710038171003817119867119905 minus 119867119905minus1100381710038171003817100381721205902 + 119887 (8)

where 119886119895 (119886119895 gt 0) is a Lagrange multiplier identified bySVM training Similarly 119862119862119905+1 is the predictive concessionamplitude value of 119860119862 in round t+1


Buyer PA MA

Order InformationRequest

Inquire

Reply

ImproveSelection

Order ProcessingAnnounce

BidReject

AdjustResult

Return

Supplier Evaluation

AcceptRefuseconfirm

Report

CA Supplier

Order Evaluation 1

Order Evaluation 2

Report

Figure 2 MAS sequence diagram of negotiation model

(3) Using the RMSE as a filter criterion as (9) we selectthe corresponding first 119896 sublearning machines

119864119894119895 = radicsum119896119894=1 (119888119894119895 minus 119862119894119895)2119896 (9)

where 119888119894119895 is the next predictive concession value in issue j ofsublearning machine i and 119862119894119895 means the actual concessionamplitude

(4) Calculate the combined weight of each submodelAccording to the RMSE value 119864119894119895 of the 119894-th submodel theweight of the submodel is obtained

120572119894 = (11198642119894119895)(sum119896119894=1 (11198642119894119895)) (10)

When all the k sublearning machines are successfully trainedselect the 119896 sublearning models with the smallest error Input

the actual concession 119862119894119895 and then get the output of ultimateconcession about issue j in the round t+1

119862119862119875119905+1119895 = 119896sum119894=1

120572119894119862119894119895 (11)

53 Utility Optimization Taking 119860119875 as an example theutility difference of sequential negotiations is used to decidewhether to stop the current consultation 119862119875119905+1119895 means apredictive concession value about issue j in round t+1 119875119875119905119895 isan actual value of buyer 119860119875 about issue j in round t

119880t = 119899sum119895=1

119908119895119875119875119905119895 (12)

119875119875119905+1119895 = 119875119875119905119895 + 119862119875119905+1119895 (13)

The error between the predictive utility value in round t+1and actual utility value in round t can be calculated bycoordinating equations (12) and (13) While Δ119880119905+1119905 gt 0 the


Start

Negotiationparameters

K-means investigation

SVM evaluation

RMSE filter

Calculate thecombined weights

Utility optimization

Common-neighboralgorithm selection

End

Yes

No

Figure 3 Flowchart of Self-adaptive negotiation optimizationstrategy

utility of concession has not been maximized it will increaseConversely end the concession

54 Selection of the Most Appropriate Partner After thenegotiation the common-neighbor algorithm [19] is appliedto compute the similarity of the issues and 119860P choose moresuitable partners according to the similarity

119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)

119863119875119862 means the total issue difference between 119860P and119860119862 119868119875 cap 119868119862 is the quantity of accredited issue after thenegotiation

Procedures are as follows (see Figure 3) First K-meanssearch was adopted to generate sample sets Second theSVM was used to learn the concession amplitude in eachevaluation sample and then eliminated the poor perfor-mance of sublearning model with RMSE and calculatedthe combined weight and the final dynamic selective SVMmodel was established Third the utility function was usedto decide whether to terminate the negotiation Finally themost appropriate partner was selected on the basis of issuesrsquosimilarity calculated with common-neighbor algorithm

Furthermore the self-adaptive negotiation optimizationstrategy is also suitable for complicated problems of bigdata in massively parallel environments The complexity ofbig data could be decreased by data processing algorithmsrsquoapplication

6 Simulation Example

Relying on modern logistics network system Yiwu hasbecome the largest small commodity distribution center inthe world The merchandise is sold to Europe America theMiddle East and South Asia and other regions Yiwu marketnow hasmore than 43million square meters of business area63 thousand operators and more than 400 thousand kinds ofproducts In 2016 the trading volume of commodity marketsreached 373 billion RMB and the total export-import volumeextended to 223 billion RMB (Yiwu China CommoditiesCity Group Official Website 2017) Yiwu Global Purchasing(wwwyiwuokcom) as an e-commerce platform contributed60 of the first value The key link of supply chain synergismis to utilize e-commerce platform services to develop ahealthy relationship of trust among partners and establishan effective mechanism for information collaboration Thispaper takes Yiwu Small Commodity Industry Cluster (SCIC)as an instance and grabs five main parameters product pricequantity delivery time warranty time and defective rateas the negotiation issue The effectiveness of self-adaptiveIntegrated Optimization Strategy (IOS) is verified by usingMatlab R2014a which is comparedwith theGeneral LearningStrategies (GLS) based on single SVM

According to the historical data analysis of electricappliances industry in Yiwu SCIC the supplier cares moreabout price quantity and delivery time while concentratingless on warranty time and defective rate The purchasingenterprise is a little bit different they focus on defectiverate rather than warranty time demonstrated detailedly inTable 3 Initial experimental datasets could be extractedfrom Dataverse repository The whole examinations wereperformed on a laptop (4GB of RAM that operated underWindows 10 desktop Intel core i3 CPU 254GHz) Inaddition we selected the open source libraries VLFeat forK-means clustering and LIBSVM for SVM algorithm withexcellent interfaces in Matlab for ease of use To get thegeneration of optimal solutions the experimental time islimited to 2 minutes

A separating hyperplane of datasets illustrated by the IOSis exhibited in Figure 4 In place of the smaller margin thehyperplane creates sheltered subregions to make most exam-ples with identical class label drop on the same side of thedecision boundary And subregions are produced by decisionboundarywith diverse piecewise shapes such as jutting out aspeninsulas that are virtually surrounded by the antagonistsThe misclassifications might comprise some stray examplessubmerged in the opponents As the crucial target of sustain-ing the native classrsquo membership the IOS eliminates the strayexamplesmdashthose characterized as black solid symbolsmdashfromthe hyperplane As mentioned above we are working on theassumption that the margin shrinkage is a price to trade offwith the misclassification decrease in the practice stage


Table 3 Intervals and weights of negotiation issue

Parameters Intervals of supplierrsquos issue Intervals of purchaserrsquos issue Weight vector of supplier Weight vector of purchaserPriceYuan [100 150] [100 130] 040 035Quantity [800 1000] [850 1200] 025 030Delivery timeMonth [15 2] [1 2] 020 020Warranty timeMonth [12 18] [15 24] 010 005Defective rate [80 95] [9095] 005 010

Table 4 Error rate () comparison of experimental results

Strategy Min error Max error Median error Average error Standard DeviationGLS 28 321 142 1586 837IOS 32 261 101 1197 605

2

1

0

minus1

minus2

210minus1minus2minus3 3

Figure 4 A separating hyperplane depicted by the IOS

Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS

Figure 5 Simulation results of error rate of 2 strategies

50 couples are selected in the experiment to predict themargin of opponent concessions by comparing two strategiesAccording to Figure 5 we can draw the conclusion that inmost cases this IOS infers lower error rate than the ordinary

Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090

Average Joint Utility Value

GLSIOS

Figure 6 The comparison of joint utility and successful agent

single SVMAdditionally the basic descriptive statistics of thedata is provided in Table 4The average error of IOS for all 50objects is 1197 with a standard deviation of 605 It couldbe seen that the IOS outperforms the GLS in four vital errormeasures The max error is 60 lower the median error is41 lower the average error is 389 lower and standarddeviation is 232 lower than the GLS respectively

In Figure 6 the average joint utility value founded byIOS is mainly concentrated in [050 075] while anothervalue is mainly concentrated in [040 070]The total averagejoint utility of the former is 0641 and 60 of agents arehigher than that value Nevertheless the numbers of the lattercalculated severally are 0565 and 46Distinctly the strategyproposed by this paper is superior toGLS in both the amountsof successful agents and joint utility value

7 Conclusions

Previous studies proposed a number of basic supply chainmodels which are difficult to spread to universal problemfields owing to the uncertainty and complexity in real-world negotiation The most fascinating modern applicationof ensemble systems lies in processing high dimensionalcomplex and big data that cannot be analyzed efficiently


by single-model methods To better solve the conflict innegotiation this paper has discussed the negotiation problemof collaborative procurement operating onMASmodelwith anegotiation optimization strategy We exploited supply chainanalysis minutely based on agent technology and machinelearning which provides a new perspective for the analysis ofintelligent SCM Apparently we perceive that the negotiationand learning are key aspects in the systemperformance by thesimulation of the proposed MAS model for the procurementmanagement of CSCThe agents have symmetric preferencescomplicating the negotiation However the learning helpedeach one acquire the ultimate strategy choice The experi-mental results show that the IOS based on dynamic selectiveensemble SVM can reduce the error rate and elevate the jointutility compared with GLS of the ordinary single learningmachine The test reveals that the model plays a key role innegotiation issue inside the intelligent SCM and the agentnegotiation performance and efficiency can be enhanced viathe combination of the improved data mining techniques

The procurement management of supply chain involvesfabrication inventory distribution and other issues and thesupply chain needs collaboration of upstream and down-stream enterprises to achieve a synergistic dynamic andtimely supply-production-marketing operationmode Futureresearch will focus on the resolution of conflict in self-adaptive negotiation to further improve the intelligent levelof supply chain

Data Availability

The datasets analyzed during the current study are availableinDataverse repository httpsdataverseharvardedudatasetxhtmlpersistentId=doi3A1079102FDVN2FVT2AQJampversion=DRAFT

Conflicts of Interest

The authors declare that there are no conflicts of interestregarding the publication of this paper

Acknowledgments

This work is supported by NSFC-Zhejiang Joint Fund forthe Integration of Industrialization and Informatization(U1509220)

References

[1] J Hallikas I Karvonen U Pulkkinen V-MVirolainen andMTuominen ldquoRisk management processes in supplier networksrdquoInternational Journal of Production Economics vol 90 no 1 pp47ndash58 2004

[2] Y Zhang L Wang and J Gao ldquoSupplier collaboration andspeed-to-market of new products the mediating and moder-ating effectsrdquo Journal of Intelligent Manufacturing vol 28 no 3pp 805ndash818 2017

[3] F Jolai J Razmi andNK Rostami ldquoA fuzzy goal programmingand meta heuristic algorithms for solving integrated produc-tion distribution planning problemrdquo Central European Journalof Operations Research vol 19 no 4 pp 547ndash569 2011

[4] Y-I Lin Y-W Chou J-Y Shiau and C-H Chu ldquoMulti-agentnegotiation based on price schedules algorithm for distributedcollaborative designrdquo Journal of Intelligent Manufacturing vol24 no 3 pp 545ndash557 2013

[5] K Govindan A Diabat and M N Popiuc ldquoContract analysisa performance measures and profit evaluation within two-echelon supply chainsrdquoComputersamp Industrial Engineering vol63 no 1 pp 58ndash74 2012

[6] M Leng and M Parlar ldquoGame theoretic applications in supplychain management a reviewrdquo Infor Information Systems ampOperational Research vol 43 no 3 pp 187ndash220 2005

[7] J-C Hennet and S Mahjoub ldquoToward the fair sharing of profitin a supply network formationrdquo International Journal of Produc-tion Economics vol 127 no 1 pp 112ndash120 2010

[8] N C Karunatillake N R Jennings I Rahwan and P McBur-ney ldquoDialogue games that agents playwithin a societyrdquoArtificialIntelligence vol 173 no 9-10 pp 935ndash981 2009

[9] Y Wu and J Angelis ldquoAchieving agility of supply chain man-agement through information technology applicationsrdquo Inter-national Federation for Information Processing vol 246 pp 245ndash253 2007

[10] O Kwon G P Im and K C Lee ldquoMACE-SCM a multi-agentand case-based reasoning collaboration mechanism for supplychain management under supply and demand uncertaintiesrdquoExpert Systems with Applications vol 33 no 3 pp 690ndash7052007

[11] F-R Lin and Y-Y Lin ldquoIntegrating multi-agent negotiation toresolve constraints in fulfilling supply chain ordersrdquo ElectronicCommerce Research and Applications vol 5 no 4 pp 313ndash3222006

[12] B Behdani A Adhitya Z Lukszo and R Srinivasan ldquoNego-tiation-based approach for order acceptance in amultiplant spe-cialty chemical manufacturing enterpriserdquo Industrial amp Engi-neering Chemistry Research vol 50 no 9 pp 5086ndash5098 2011

[13] J Zhang F Ren andMZhang ldquoBayesian-basedpreference pre-diction in bilateral multi-issue negotiation between intelligentagentsrdquo Knowledge-Based Systems vol 84 pp 108ndash120 2015

[14] L Chen H Dong and Y Zhou ldquoA reinforcement learning opti-mized negotiation method based on mediator agentrdquo ExpertSystems with Applications vol 41 no 16 pp 7630ndash7640 2014

[15] Z Ma C Wang Y Niu X Wang and L Shen ldquoA saliency-based reinforcement learning approach for a UAV to avoidflying obstaclesrdquoRobotics and Autonomous Systems vol 100 pp108ndash118 2018

[16] J Heinermann and O Kramer ldquoMachine learning ensemblesfor wind power predictionrdquo Journal of Renewable Energy vol89 pp 671ndash679 2016

[17] Y Liu B He D Dong et al ldquoParticle swarm optimizationbased selective ensemble of online sequential extreme learningmachinerdquo Mathematical Problems in Engineering vol 2015Article ID 504120 10 pages 2015

[18] N B Peng Y X Zhang and Y H Zhao ldquoA SVM-kNNmethodfor quasar-star classificationrdquo Science China Physics Mechanicsamp Astronomy vol 56 no 6 pp 1227ndash1234 2013

[19] Y H He D B Chen et al ldquoSimilarity algorithm based on userscommon neighbors and grade informationrdquo Computer Sciencevol 37 no 9 pp 184ndash186 2010

Hindawiwwwhindawicom Volume 2018

MathematicsJournal of


Mathematical Problems in Engineering

Applied MathematicsJournal of


Probability and StatisticsHindawiwwwhindawicom Volume 2018

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of



Engineering Mathematics

International Journal of


Operations ResearchAdvances in

Journal of


Function SpacesAbstract and Applied AnalysisHindawiwwwhindawicom Volume 2018

International Journal of Mathematics and Mathematical Sciences


Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Hindawiwwwhindawicom Volume 2018Volume 2018

Numerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisNumerical AnalysisAdvances inAdvances in Discrete Dynamics in

Nature and SocietyHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Dierential EquationsInternational Journal of

Volume 2018


Decision SciencesAdvances in


AnalysisInternational Journal of


Stochastic AnalysisInternational Journal of

Submit your manuscripts atwwwhindawicom


as a basis to formulate autonomous negotiation Howeverthose approaches are considered to be complicated to spreadto widespread problem fields due to the uncertainty andcomplexity in real-world negotiation This paper amelioratesthe negotiation model combining multi-agent system (MAS)with machine learning for further tackling the conflict incooperation New model provides a buyer with a method forpurchasing a product systematically And it helps in achievinga win-win cooperation between two sides during the processof collaborative procurement with supply chain as far asfeasible

The remainder of the paper is structured as follows Sec-tion 2 shows literature review Section 3 recalls some generalconcepts of key techniques Section 4 is devoted to a nego-tiation model of collaborative procurement based on MASSection 5 describes a self-adaptive negotiation optimizationstrategy combined with dynamic selective ensemble learningFinally the experiments design results and concludingremarks are presented in Sections 6 and 7 respectively

2 Literature Review

Given that research on negotiation of collaborative pro-curement is new and largely fragmented it is practicallyparamount to arouse individualsrsquo attention Previous studieshave nevertheless proposed a basic model of supply chainsand a negotiation strategy for solving conflicts in consider-ation of efficiency and cost A multi-objective cooperativeproductionndashdistribution planning model was formulated byJolai et al [3] applying the fuzzy goal programming approachto maximize the gains of all participators To discover theoptimal solutions of resource allocation Lin et al [4] rec-ommended a collaborative negotiation mechanism that wasbuilt on price schedules decomposition algorithm But thepopular methods to research the negotiationrsquos conundrumin SCM involve Game Theory and artificial intelligence(AI) Game theorists deem the negotiation as an incompletedynamic information game and attempt to settle the gameby offering some predictions on certain conditions [5]Primary methodological tools of Game Theory are Nashgame [6] and Stackelberg game [7] which concentrate onthe sequential and simultaneous decision-making of multipleplayers respectively For those relevant analytic modelingstudies the problem is analyzed mostly from a theoreticalperspective Despite being extremely successful in a quantityof situations the game theoretical approach is considered tobe difficult to spread to universal problem fields owing to theuncertainty and complexity in real-world negotiation

Compared to Game Theory participants that bargainwith consideration of human preference and thoughts couldbe considerably represented by the agent technology whichis a branch of AI [8] The use of information and communi-cation technology tools offering the capacities of customersensitivity information sharing and process integration isobserved as the uppermost enabler for this collaborative per-ceptionrsquos realization [9] In computer science an agent is gen-erally considered as a software entity which is autonomous tocommunicate and coordinatewith other agents to accomplishits design objectives Consequently multi-agent simulation

modeling which originated from AI is suitable for theconduction of distributed system and has certain advantagesin being testable quantifiable and efficient It is superior inexpansibility is easy to configure and has been widely usedin the SCM Kwon et al [10] constructed an integrated frame-work that was based on multi-agent cooperation and case-based reasoning to help address emerging uncertainties Linet al [11] demonstrated a supply chain coordination modelof multi-agent and put forward a conflict solution methodbuilt on constraint satisfaction algorithm due to the differentformof demand Considering the conflict between businessescaused by the difference of information asymmetry and goalsBehdani et al [12] developed a negotiation method based onmulti-agent in the condition that demand is uncertain Thesignificance of addressing negotiation mechanisms for col-laborative matters is shown by the discussed literatures Thecombination of negotiation model and optimization technol-ogy is requisite to help negotiators achieve optimal selections

In order to better promote the agentrsquos self-adaptive nego-tiation ability an army of scholars have begun to introducemachine learning into the negotiation Bayesian Learningestimates the probability distribution of opponent negotia-tion parameters and preferences and adaptively adjusts theconcession strategy [13] Q-Learning generates the optimalnegotiation strategy by calculating the utility cumulativevalue [14] Radial Basis Function (RBF) neural network iscapable of optimizing the Actor-Critic learning algorithm topredict and amend the concession magnitude of agents [15]Unfortunately previous self-adaptive negotiation is built on asingle or integrated learning machine to draw the final result[16] Selective ensemble learning improves the efficiency ofgeneral integrated learning machine by eliminating the lessaccurate ones in sublearning models [17]

This paper is built on our previous work in the fieldof automated negotiation In particular it lays the foun-dation for accomplishing an experiment to investigate theperformance of agent which is operating in the supply chainsystem and equipped with our negotiation model The maincontribution consists of constructing a negotiation modelconcerning collaborative procurement based on MAS byanalyzing the characteristics of multilateral transact andproposing a negotiation strategy founded on dynamic selec-tive ensemble learning We exploited supply chain analysisdetailedly that was based on agent technology which detectsnovel patterns through the improved data mining techniquesand provides a new perspective for the analysis of intelligentSCMMoreover agent jobwas led by this association betweenintelligent agents and machine learning to do faster andbetter And the negotiation strategy has also potential for bigdata decrement and compression

3 Methods

31 Machine Learning Machine learning gradually becomesan irreplaceable method for processing data in the big dataera As an embranchment of AI it has entered foreland ofthe mainstream computer sciencersquos research that often usesstatistical techniques to give agents the ability to learn withdata without being explicitly programmedMachine learning













NM = A PwU (1)



119875119863 (119875119902 119875119894) = radicsum119894isin119871

(119875119902 minus 119875119894)2 (2)














Δ119863119905 = 10038161003816100381610038161003816119875119862119905 minus 119875119875119905 10038161003816100381610038161003816 (3)

119862119875119905 = 119905sum119894=2

Δ119863119894119875119875119894minus1 (4)

119862119875119905+1 = 119908119879 lowast (120593 (119905) 120593 (Δ119863119905) 120593 (119862119875119905 ) 120593 (119862119862119905 )) + 119887 (5)



max 0 10038161003816100381610038161003816119910 minus 119862119875119905+110038161003816100381610038161003816 minus 120576 (6)

119862119875t+1 = 119899sum119895=1

(119886119895 minus 1198861015840j )119870 (119867119905 119867119905minus1) + 119887 (7)

119862119875t+1 = 119899sum119895=1




Buyer PA MA


Inquire

Reply

ImproveSelection


BidReject

AdjustResult

Return

Supplier Evaluation

AcceptRefuseconfirm

Report

CA Supplier

Order Evaluation 1

Order Evaluation 2

Report



119864119894119895 = radicsum119896119894=1 (119888119894119895 minus 119862119894119895)2119896 (9)



120572119894 = (11198642119894119895)(sum119896119894=1 (11198642119894119895)) (10)



119862119862119875119905+1119895 = 119896sum119894=1

120572119894119862119894119895 (11)


119880t = 119899sum119895=1

119908119895119875119875119905119895 (12)

119875119875119905+1119895 = 119875119875119905119895 + 119862119875119905+1119895 (13)



Start



SVM evaluation

RMSE filter




End

Yes

No




119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018




















NM = A PwU (1)



119875119863 (119875119902 119875119894) = radicsum119894isin119871

(119875119902 minus 119875119894)2 (2)














Δ119863119905 = 10038161003816100381610038161003816119875119862119905 minus 119875119875119905 10038161003816100381610038161003816 (3)

119862119875119905 = 119905sum119894=2

Δ119863119894119875119875119894minus1 (4)

119862119875119905+1 = 119908119879 lowast (120593 (119905) 120593 (Δ119863119905) 120593 (119862119875119905 ) 120593 (119862119862119905 )) + 119887 (5)



max 0 10038161003816100381610038161003816119910 minus 119862119875119905+110038161003816100381610038161003816 minus 120576 (6)

119862119875t+1 = 119899sum119895=1

(119886119895 minus 1198861015840j )119870 (119867119905 119867119905minus1) + 119887 (7)

119862119875t+1 = 119899sum119895=1




Buyer PA MA


Inquire

Reply

ImproveSelection


BidReject

AdjustResult

Return

Supplier Evaluation

AcceptRefuseconfirm

Report

CA Supplier

Order Evaluation 1

Order Evaluation 2

Report



119864119894119895 = radicsum119896119894=1 (119888119894119895 minus 119862119894119895)2119896 (9)



120572119894 = (11198642119894119895)(sum119896119894=1 (11198642119894119895)) (10)



119862119862119875119905+1119895 = 119896sum119894=1

120572119894119862119894119895 (11)


119880t = 119899sum119895=1

119908119895119875119875119905119895 (12)

119875119875119905+1119895 = 119875119875119905119895 + 119862119875119905+1119895 (13)



Start



SVM evaluation

RMSE filter




End

Yes

No




119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018




















Δ119863119905 = 10038161003816100381610038161003816119875119862119905 minus 119875119875119905 10038161003816100381610038161003816 (3)

119862119875119905 = 119905sum119894=2

Δ119863119894119875119875119894minus1 (4)

119862119875119905+1 = 119908119879 lowast (120593 (119905) 120593 (Δ119863119905) 120593 (119862119875119905 ) 120593 (119862119862119905 )) + 119887 (5)



max 0 10038161003816100381610038161003816119910 minus 119862119875119905+110038161003816100381610038161003816 minus 120576 (6)

119862119875t+1 = 119899sum119895=1

(119886119895 minus 1198861015840j )119870 (119867119905 119867119905minus1) + 119887 (7)

119862119875t+1 = 119899sum119895=1




Buyer PA MA


Inquire

Reply

ImproveSelection


BidReject

AdjustResult

Return

Supplier Evaluation

AcceptRefuseconfirm

Report

CA Supplier

Order Evaluation 1

Order Evaluation 2

Report



119864119894119895 = radicsum119896119894=1 (119888119894119895 minus 119862119894119895)2119896 (9)



120572119894 = (11198642119894119895)(sum119896119894=1 (11198642119894119895)) (10)



119862119862119875119905+1119895 = 119896sum119894=1

120572119894119862119894119895 (11)


119880t = 119899sum119895=1

119908119895119875119875119905119895 (12)

119875119875119905+1119895 = 119875119875119905119895 + 119862119875119905+1119895 (13)



Start



SVM evaluation

RMSE filter




End

Yes

No




119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018









Buyer PA MA


Inquire

Reply

ImproveSelection


BidReject

AdjustResult

Return

Supplier Evaluation

AcceptRefuseconfirm

Report

CA Supplier

Order Evaluation 1

Order Evaluation 2

Report



119864119894119895 = radicsum119896119894=1 (119888119894119895 minus 119862119894119895)2119896 (9)



120572119894 = (11198642119894119895)(sum119896119894=1 (11198642119894119895)) (10)



119862119862119875119905+1119895 = 119896sum119894=1

120572119894119862119894119895 (11)


119880t = 119899sum119895=1

119908119895119875119875119905119895 (12)

119875119875119905+1119895 = 119875119875119905119895 + 119862119875119905+1119895 (13)



Start



SVM evaluation

RMSE filter




End

Yes

No




119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018









Start



SVM evaluation

RMSE filter




End

Yes

No




119878119875119862 = (1 + 119890minus1198631198751198622) lowast 1003817100381710038171003817119868119875 cap 1198681198621003817100381710038171003817 (14)













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018













2

1

0

minus1

minus2



Erro

r Rat

e (

)

Number of Samples

35

30

25

20

15

10

5

0

1 6 11 16 21 26 31 36 41 46

IOSGLS



Num

ber o

f Age

nts

12

10

8

6

4

2

0

035

040

045

050

055

060

065

070

075

08

0

085

090


GLSIOS




7 Conclusions





Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018











Data Availability




Acknowledgments


References



























Journal of












Journal of







Volume 2018






Volume 2018















Journal of












Journal of







Volume 2018






Volume 2018








Documents

A Negotiation Optimization Strategy of Collaborative