=Paper=
{{Paper
|id=Vol-3026/paper23
|storemode=property
|title=Broker-Based Trade Allocation in Agent-Based E-commerce
|pdfUrl=https://ceur-ws.org/Vol-3026/paper23.pdf
|volume=Vol-3026
|authors=Dien Tuan Le,Dang My Phuong Phan,Van Loc Tran
}}
==Broker-Based Trade Allocation in Agent-Based E-commerce==
<pdf width="1500px">https://ceur-ws.org/Vol-3026/paper23.pdf</pdf>
<pre>
                Broker-Based Trade Allocation in Agent-
                        Based E-commerce

       Dien Tuan Le**[0000-0001-8475-5096], Dang My Phuong Phan and Van Loc Tran

                 University of Economics, The University of Danang, Vietnam
                  {ledientuan,phuong.pdm,loctv}@due.edu.vn



        Abstract. Online business environment represents complex systems where the
        business transactions between buyers and sellers are carried out based on multiple
        constraints from buyers and sellers. The complexity of the online business envi-
        ronment between buyers and sellers has encouraged the use of modelling broker
        to solve allocations. Thus, this paper proposes a novel approach for a broker agent
        to allocate buyers’ requirements to sellers’ offers using the method of artificial
        intelligence. The major contributions of this paper are that (i) a proposed frame-
        work for a broker agent is divided in four stages: receiving, calculation, filtering
        and allocating; (ii) a novel calculation method is to calculate buyers’ satisfaction
        degree as per sellers’ offers to determine a constraint satisfaction layer; and (iii)
        CSP model for trade allocation is built to help a broker agent to find an optimal
        allocation solution to satisfy buyers and sellers’ various preferential require-
        ments. Finally, the results of the simulation experiment demonstrate the proposed
        approach is flexible and effective under the consideration of different constraints.

        Keywords: Buyers’ satisfaction degree, Allocation process, Multi-attribute
        trading.


 1      Introduction

    E-commerce has recently made remarkable achievements in the era of globalization
 together with rapid technological changes. Especially, the emergence of artificial intel-
 ligent technology makes e-commerce systems more intelligent so it is convenient for
 buyers and sellers to carry out business transactions via brokers in digital business en-
 vironments [1-2]. A broker is responsible to receive requirements of buyers and sellers
 and to implement trade allocation procedures with regard to the consideration of differ-
 ent situations in complex business environments [3-5]. Research on brokers or


  Copyright © by the paper’s authors. Use permitted under Creative Commons License Attribu-
   tion 4.0 International (CC BY 4.0). In: N. D. Vo, O.-J. Lee, K.-H. N. Bui, H. G. Lim, H.-J.
   Jeon, P.-M. Nguyen, B. Q. Tuyen, J.-T. Kim, J. J. Jung, T. A. Vo (eds.): Proceedings of the
   2nd International Conference on Human-centered Artificial Intelligence (Computing4Human
   2021), Da Nang, Viet Nam, 28-October-2021, published at http://ceur-ws.org
** Corresponding author.
210 Le et al.


intermediaries as the third party of the trading processes in e-commerce has been active
directions in recent years. Jiang et al. [6] proposed an optimization approach to solve
matching problems through a broker using a single objective function under the con-
sideration of estimated prices. A matching model for a broker is built to maximize the
weight sum of the sellers’ and buyers’ satisfaction degree based on requirements of
buyers and sellers. Li et al. [7] proposed a new approach to allocate buyers to sellers
using a multi-objective optimization model through a broker. Furthermore, an agent-
based framework was built to carry out matching processes in three layers: the interface
layer, the matching layer and the database layer. Jiang et al. [8] built an optimal match-
ing approach for brokers under the consideration of multi-attribute trading with fuzzy
information and indivisible demand considerations.
   As mentioned above, such approaches have focused on studying brokers as the third
party to allocate buyers’ requirements to sellers’ offers in business environments. How-
ever, there are few theories and guidelines to help a broker agent in intelligent e-com-
merce systems to find out optimal allocation solutions using CSP techniques and multi-
attribute trading. Most current brokers in e-marketplace only provide buyers’ or sellers’
trade information and do not really employ artificial intelligence techniques to allocate
buyers to sellers. The lack of a comprehensive optimization approach on allocation
could not provide a solid foundation to improve market efficiency in digital business
environments. Therefore, how to allocate buyers to sellers using CSP techniques under
the consideration of constraints between buyers and sellers is one of the most important
challenges for a broker agent. The novel approach in this paper is proposed to solve this
challenge.
   The rest of this paper is organized as follows. Problem description is presented in
Section 2. The proposed approach for a broker agent is introduced in Section 3. An
experiment is described in Section 4. Section 5 gives conclusions and points out our
future work.


2       Problem description

There are three members in the trading process with multi-attribute trading, i.e., buyers,
sellers and a broker. The broker is often called a facilitator, who acts as an intermediary
between buyers and sellers in multi-attribute trading. The consideration of the seller-
buyer matching problem through a broker agent is shown in Fig. 1.

                            Requirements                  Offers

                   Buyers                                              Sellers
                                            Broker
                           Matching results Agent Matching results




                Fig. 1. The seller-buyer matching problem through a broker agent

   The broker agent’s responsibility is to match 𝑛 (𝑛 ≥ 1) buyers with 𝑚 (𝑚 ≥ 1)
sellers for the same commodity with multi-attribute trading in order to satisfy buyer’s
                                     Broker-Based Trade Allocation in Agent-Based E-commerce 211


requirements and it is assumed that each buyer (seller) can buy (sell) commodities for
each seller (buyer) at most.
   Let a set of buyers 𝑩 = {𝐵1 , 𝐵2 , … , 𝐵𝑛 }, a set of sellers 𝑺 = {𝑆1 , 𝑆2 , … , 𝑆𝑚 }, buyers’
requirements and sellers’ offers are related to multi attributes which are divided to at-
tributes with hard constraints and soft constraints [9]. Buyers’ requirements with multi-
attributes are split into a set of attributes with hard constraints 𝑯 = {ℎ1 , ℎ2 , … , ℎ𝑧 } and
set of attributes with soft constraints 𝑨 = {𝑎1 , 𝑎2 , … , 𝑎𝑘 }. Let 𝑛 be number of buyers,
𝑖 ∈ [1,2 … 𝑛] is the index of 𝐵𝑖 ’s requirements and 𝐶𝑖𝑖 ′ (𝑖 ′ ∈ 𝑧 + 𝑘)) is the crisp value
of the 𝑖 ′𝑡ℎ attribute of 𝐵𝑖 ’s requirements. Similarly, let m be number of buyers, 𝑗 ∈
[1,2 … 𝑚] is the index of 𝑆𝑗 ’s offers and 𝑄𝑗𝑖 ′ (𝑖 ′ ∈ 𝑧 + 𝑘)) is the crisp value of the 𝑖 ′𝑡ℎ
attribute of 𝑆𝑗 ’s offers.
   Based on the above analysis, the key problem is how to help a broker agent to
achieve the optimal allocation results between trading (buyers and sellers) agents in
agent-based e-commerce using CPS model. Therefore, the proposed allocation ap-
proach for a broker agent can solve this problem and is presented in Section 3.


3         A proposed approach for a broker agent

3.1       Framework of the proposed approach
The framework of the proposed approach for a broker agent to allocate buyers’ require-
ments to sellers’ offers using constraint satisfaction problem models is presented in Fig.
2 as follows.

                                                           Constraint satisfaction layer
 Receive buyers requirements and sellers offers
                                                                    The group of buyers
        Calculate buyers satisfaction degree to
        determine constraint satisfaction layer

                                                                    The group of sellers
      Filter constraint satisfaction layer based on
               buyers satisfaction degree

    Allocate buyer s requirements to seller s offers    • Constraint satisfaction problem models


               Obtain allocation results

                Fig. 2. The framework of proposed approach for a broker agent

      In the framework, trading information related to buyers’ requirements and sellers’
offers is submitted to a broker agent. Then, buyers’ requirements are classified into
attributes with three kinds of constraints, i.e., attributes with hard constraints, benefit
soft constraints, and cost soft constraints. Based on attribute classification, buyers’ sat-
isfaction degree for each type of attribute is calculated as per sellers’ offers to determine
constraint satisfaction layer. After that, a broker agent can filter a constraint satisfaction
degree layer based on buyers’ satisfaction degree to achieve a broker agent’s goals.
212 Le et al.


Finally, a broker agent carries out trade allocation using constraint satisfaction problem
models to obtain the optimal allocation results to satisfy buyers’ requirements as per
sellers’ offers.


3.2       Calculation of buyers’ satisfaction degree based on sellers’ offers.
Based on notations of buyers’ requirements and sellers’ offers in Section 2, the calcu-
lation method of buyers’ satisfaction degree for all attributes is presented as follows:
    (i) for an attribute type with hard constraints:

  ℎ𝑔      -1        if Cig ≠Qjg
𝛽𝑖𝑗 = {                                                                                    (1)
            1 if Cig=Qjg
  ℎ𝑔                                                                                     𝑎𝑔
𝛽𝑖𝑗 = -1 means that 𝐵𝑖 does not match with 𝑆𝑗 for attribute ℎ𝑔 (𝑔 ∈ 𝑧) and 𝛽𝑖𝑗 =1
means that 𝐵𝑖 matches with 𝑆𝑗 for attribute ℎ𝑔 .
                                                                                𝑎
   (ii) For an attribute type with benefit soft constraints: if Qjl < 𝐶il then 𝛽𝑖𝑗𝑙 =-1. It
                                                          𝑎
means that 𝐵𝑖 does not satisfy 𝑆𝑗 . If Qjl ≥ C , then 𝛽𝑖𝑗𝑙 is calculated as follows:
                                                il

            Qjl - Qmin-l +φ     t
  𝑎
𝛽𝑖𝑗𝑙 =( 𝑄                       ),                                                         (2)
          max-l - Qmin-l     +φ

              𝐶𝑖𝑙
where 𝑡 =        ⁄𝑄       , Qmin-l is the minimal value of seller in the set of values for the
                    𝑚𝑖𝑛−𝑙
attribute 𝑎𝑙 and Qmax-l is the maximal value of a seller in the set of values for the attrib-
ute 𝑎𝑙 . A value t ∈ (0,1] helps a broker agent to carry out comparing a buyer’s satisfac-
                                              𝑎
tion degree when 𝑡 is used to calculate 𝛽𝑖𝑗𝑙 . φ= Qmin-l/2, φ helps a broker agent to solve
                                                                                             𝑎
some special cases such as only one seller in e-commerce systems or Qmax-l = Qmin-l. 𝛽𝑖𝑗𝑙
                                                                                             𝑎
means that 𝐵𝑖 matches with 𝑆𝑗 for attribute 𝑎𝑙 with a buyer’s satisfaction degree 𝛽𝑖𝑗𝑙 .
  𝑎                                  𝑎
𝛽𝑖𝑗𝑙 is in-between 0 and 1. If 𝛽𝑖𝑗𝑙 is near 1, it means that 𝑆𝑗 is highly satisfied by 𝐵𝑖 for
attribute 𝑎𝑙 (𝑙 ∈ 𝑘).
                                                                              𝑎
    (iii) for an attribute type with cost soft constraints: if Cil <Qjl then 𝛽𝑖𝑗𝑙 =-1. It means
                                                 𝑎
that 𝑆𝑗 does not satisfy 𝐵𝑖 . If Cil ≥Qjl , then 𝛽𝑖𝑗𝑙 is calculated as follows:
                                1
 𝑎       Qmax-l - Qjl+ φ 𝑡
𝛽𝑖𝑗𝑙 =(                   ),                                                               (3)
        Q    -Q
            max-l      +φ
                     min-l



3.3       Applying CSP for a broker agent to carry out trade allocations
To model CSP for allocation processes, a broker agent needs to determine constraints
based on the calculation of buyers’ satisfaction degree. Relationships between buyers
and sellers through buyers’ satisfaction degree are presented in Table 1 as follows.
                             Broker-Based Trade Allocation in Agent-Based E-commerce 213


           Table 1. Buyers’ satisfaction degree as per sellers’ offers in CSP model
                  𝑆1            𝑆2               𝑆3               …                    𝑆𝑚
     𝐵1          𝛽11            𝛽12              𝛽13              …                   𝛽1𝑚
     𝐵2          𝛽21            𝛽22              𝛽23              …                   𝛽2𝑚
     …            …              …                …               …                     …
     𝐵𝑛          𝐵𝑛1            𝐵𝑛2              𝐵𝑛3              …                   𝛽𝑛𝑚

   where 𝛽𝑛𝑚 is 𝐵𝑛 ’s satisfaction degree as per 𝑆𝑚 ’s offers under the consideration of
multi-attribute trading. If 𝑆𝑚 does not satisfy 𝐵𝑛 for a certain attribute then 𝛽𝑛𝑚 = −1.
It means that 𝑆𝑚 does not satisfy 𝐵𝑛 and the constraint between 𝑆𝑚 and 𝐵𝑛 in a CSP
model is created for trade allocation processes. Trade allocation through a broker agent
between 𝑛 buyers and 𝑚 sellers under the consideration of constraints in this paper are
solved using CSP techniques. Thus, variables, domain of variables and constraints are
defined in the CSP model for trade allocation as follows.

• Variables and domain of variables
𝑥𝑖𝑗 = 1 𝑜𝑟 0, ∀𝑖 ∈ 𝑛, ∀𝑗 ∈ 𝑚                                                                (4)

• Constraints
∑𝑛𝑖=1 𝑥𝑖𝑗 ≤ 1, 𝑗 = 1,2, … , 𝑚                                                               (5)

∑𝑚
 𝑗=1 𝑥𝑖𝑗 ≤ 1, 𝑖 = 1,2, … , 𝑛                                                                (6)
            𝑎                             𝑔 𝑎
𝑥𝑖𝑗 = 0 𝑖𝑓 𝛽𝑖𝑗𝑙 = −1 (𝑙 = 1,2, … 𝑘) 𝑜𝑟 𝛽𝑖𝑗 = −1 (𝑔 = 1,2, … 𝑧)                              (7)

Constraints in Equation 4 is decision variable constraints, if buyer 𝐵𝑖 matches with
seller 𝑆𝑗 then 𝑥𝑖𝑗 = 1; otherwise 𝑥𝑖𝑗 = 0. Constraints in Equation 5 and 6 are that each
buyer (seller) only matches with each seller (buyer) maximally. Constraints in Equation
7 determine a constraint satisfaction layer. After a broker agent defines variables and
determines the domain of variables and constraints, the broker agent uses the CSP tech-
niques to find out the optimal allocation solutions to satisfy buyers’ requirements as per
sellers’ offers.


4      Experiments

In this section, experimental results demonstrate that the proposed approach for a bro-
ker agent is flexible and provides reasonable allocation solutions in business environ-
ments to support decision making. In particular, the experimental setting is presented
in subsection 4.1 and the experimental results are evaluated in subsection 4.2.
214 Le et al.


4.1     Experimental setting
In the experiments, we generate an artificial data of 10 buyers and 10 sellers in the
digital business environments of the real estate in Vietnam. Buyers’ requirements and
sellers’ offers for goods contain five attributes, i.e., location (a1), price (a2), building
size (a3), number of payment days (a4), property (a5). Based on buyers’ requirements
related to five attributes, location (a1) and property (a5) are attributes with hard con-
straints because their constraints must be satisfied while price (a2) is an attribute with
cost soft constraints and building size (a3), number of payment days (a4) are attributes
with benefit soft constraints because buyers can be accepted to relax the values of the
attributes to achieve their satisfaction degrees.
   In the experiments, the proposed allocation approach for a broker agent is employed
to find out optimal pair allocations under the different situations in the digital business
environments of real estate in Vietnam. More specifically, two representative experi-
ments are presented below to demonstrate the proposed approach effectively.


4.2     Experimental results
   Experiment 1: Trade allocations without filtering buyers’ satisfaction degree
   In this experiment, a broker agent applies CSP in Python to find out the trade allo-
cation results under the consideration of 10 buyers’ requirements and 10 sellers’ offers
related to 5 attributes and without filtering buyers’ satisfaction degree. In particular, a
broker agent calculates each buyer’s satisfaction degree based on each seller’s offers to
determine the layer of constraint satisfaction and based on each buyer’s satisfaction
degree as per each seller’s offers, a broker agent determines constraints for the CSP
model before a trade allocation process is carried out. Based on data of 10 buyers’ re-
quirements and 10 sellers’ offers for goods with 5 five attributes in subsection 4.1, a
broker agent finds out 145,421 solutions for trade allocation. The specific results with-
out filtering buyers’ satisfaction degree are presented in Fig. 3.




                Fig. 3. The number of allocation solutions using CSP in Python
                             Broker-Based Trade Allocation in Agent-Based E-commerce 215


    Based on experimental results, a broker agent considers to select which solutions
from allocation results to achieve a broker agent’s goals as per the constraints of buyers
or sellers or both. For instance, if a broker agent wants to choose an allocation solution
from 145,421 solutions to maximize buyers’ satisfaction degree, solution 128,861 of
allocation is perfect because the normalization of buyers’ satisfaction degree in this
solution equals 1. The specific allocation result in this solution is 𝐵1 ↔ 𝑆10 , 𝐵2 ↔
𝑆3 , 𝐵3 ↔ 𝑆8 , 𝐵4 ↔ 𝑆7 , 𝐵5 ↔ 𝑆9 , 𝐵6 ↔ 𝑆2 , 𝐵7 ↔ 𝑆1 , 𝐵8 ↔ 𝑆6 , 𝐵9 ↔ 𝑆5 , 𝐵10 ↔ 𝑆4 .
    Experiment 2: The consideration of relationship between buyers’ satisfaction de-
gree and a number of allocation solutions
       In the second experiment, the relationship between buyers’ satisfaction degree
and a number of allocation solutions was studied. Buyers’ satisfaction degree is set to
increase gradually to find out a number of allocation solutions. In particular, 8 levels of
buyers’ satisfaction degree are set for this experiment, and a number of allocation so-
lutions as per each level of buyers’ satisfaction degree is found. The specific results are
presented in Fig. 4. Based on allocation results in Fig. 4, it is clear that if the require-
ment of buyers’ satisfaction degree is high, the number of allocation solutions is low
and vice versa. It demonstrates that the proposed approach works effectively. In sum-
mary, a broker agent can take off flexible allocation solutions to satisfy buyers’ require-
ments as per sellers’ offers based on the constraints of buyers’ satisfaction degree.




   Fig. 4. The relationship between buyers’ satisfaction degree and a number of alloca-
tion solutions


5      Conclusion and Future work

     This paper proposes the allocation approach for a broker agent in agent-based e-
commerce using the CSP model. The proposed approach is novel because (i) the design
of allocation approach for a broker agent is general so it can be applied into broad do-
mains such as real estate, banking, education, etc.; (ii) the formula system is built to
calculate buyers’ satisfaction degree as per sellers’ offer for multi-attribute trading; and
(iii) a broker agent’s allocation process is implemented based on the constraint satis-
faction process using CSP. CSP helps a broker agent to find out an optimal solution by
216 Le et al.


means of satisfying various preferences of buyers and sellers. The experimental results
demonstrate the promising performance of the proposed approach in aspects of satisfy-
ing buyers’ requirements as per sellers’ offers and finding out optimal allocation results
based on constraints. Furthermore, future research would extend the proposed approach
to solve complex business environments such as the consideration of negotiation, price
policies, etc.


References
 1. Tang, J.: Artificial intelligence-based e-commerce platform based on SaaS and neural net-
    works. In: the Fourth International Conference on Inventive Systems and Control (ICISC),
    pp. 421-424, IEEE, India (2020). https://doi.org/10.1109/ICISC47916.2020.9171193
 2. Wang, X., Zhang, M., Ren, F., Ito, T.: GongBroker: A broker model for power trading in
    smart grid markets. In: International Conference on Web Intelligent and Intelligent Agent
    Technology (WI-IAT), pp. 21-24, IEEE, Singapore (2016). https://doi.org/10.1109/WI-
    IAT.2015.108
 3. Le, D. T., Zhang, M., Ren, F.: An economic model-based matching approach between buy-
    ers and sellers through a broker in an open e-marketplace. Journal of Systems Science and
    Systems Engineering, 27(2), 156-179 (2018). https://doi.org/10.1007/s11518-018-5362-z
 4. Nagarajan, R., Thirunavukarasu, R., Selvamuthukumaran, S.: A cloud broker framework for
    infrastructure service discovery using semantic network. International Journal of Intelligent
    Engineering Systems, 11(3), 11-19 (2018). http://www.inass.org/2018/2018063002.pdf
 5. Zhang, X., Yuan, X. M., Yuan, J. G., Hu, H. J.: Multi-objective approach for broker domi-
    nant supply-demand matching decision. Journal of Interdisciplinary Mathematics, 20(4),
    1139-1152 (2017). https://doi.org/10.1080/09720502.2017.1358887
 6. Jiang, Y. P., Fan, Z. P., Liang, H. M., Sun, M.: An optimization approach for existing home
    seller-buyer matching. Journal of the Operational Research Society, 70(2), 237-254 (2019).
    https://doi.org/10.1080/01605682.2018.1427432
 7. Li, Y. H., Fan, Z. P., Chen, X., Kang, F.: An Agent-Based Framework for Matching Buy-
    ers and Sellers. In: the 4th International Conference on Wireless Communications, Net-
    working and Mobile Computing, pp. 1-4, IEEE, China (2008).
    https://doi.org/10.1109/WiCom.2008.2224
 8. Jiang, Z. Z., Shu, P., Ip, W. H., Chen, X.: Fuzzy multi objective modeling and optimization
    for one-shot multi attribute exchanges with indivisible demand. IEEE Transactions on Fuzzy
    Systems, 24(3) 708-723 (2015). https://doi.org/10.1109/TFUZZ.2015.2476516
 9. Jiang, Z. Z., Ip, W. H., Lau, H. C., Fan, Z. P.: Multi-objective optimization matching for
    one-shot multi-attribute exchanges with quantity discounts in E-brokerage. Expert Systems
    with Applications, 38(4), 4169-4180 (2011). http://dx.doi.org/10.1016/j.eswa.2010.09.079

</pre>