Study on the E-commerce Platform Based on Artificial Intelligence

Headnote

Abstract: With the rapid development of Internet and information technology, E-commerce has been gradually accepted by people and enters our daily life. Its rapid development has promoted the integration and development of the global economy with larger opportunities and challenges. E-commerce is a new business transaction way conducted through modern information technology and computer network. Mobile Agent technology in artificial intelligence is featured with mobility, autonomy, responsibility and intellectuality, which laid a good basis for the realization of intelligent E-commerce. On account of the issues in the E-commerce negotiation process, Multi-Agent technology classified E-commerce according to the property of the business, proposed the negotiation model based on GA (Genetic Algorithm) and accomplished the overall design of the E-commerce system of Mobile Agent. Through the establishment of reasonable utility function and learning of the opposite's preferences, this design model can relatively maximize the benefits of both merchants and customers, and realize the intellectualization of the E-commerce transaction process. According to the experimental analysis, the application of E-commerce platform based on artificial intelligence in the development of E-commerce system can effectively solve the problems existed in the current E-commerce systems.

Keywords: E-commerce platform; Agent; artificial intelligence; Genetic Algorithm

(ProQuest: ... denotes formulae omitted.)

1. Introduction

In many cases, E-commerce is noted as electronic trade, electronic transaction, paperless trade or network trade, etc. Although the word "E-commerce" is widely used, E-commerce is still not mature. The form, content and characteristics of E-commerce are still evolving. Hence, it is not easy to give a strict definition of E-commerce. Many people define E-commerce according to their own understandings. For example, Tingpeng Liang (He M, Jennings N R, Leung H F., 2003) proposed the intermediate system model based on Agent E-commerce; Jeffrey (Tsvetovatyy M, Gini M, Mobasher B., 1997) et al. applied Agent technology in the web-based negotiation support system and developed the E-commerce system model MagNET based on Aglet platform; Chartree (Glushko R J, Tenenbaum J M, Meltzer B., 1999) et al. first discussed the workflow issues in E-commerce system based on Mobile Agent; Chan (Wang W, Benbasat I., 2007) et al. raised the SIAS system. U.S. expert Dr. EmmeLhainz (Schubert P, Ginsburg M., 2000) pointed out that E-commerce is to realize the coordination of supplies or personnel through internet so that to facilitate commercial exchange activities. Chinese expert and researcher Wangke defined E-commerce as an overall process of realizing commercial exchange and industrial work through electronic tools based on computer and communication perspectives. Ouyangwu, an expert of Institute of Quantitative & Technical Economics, Chinese Academy of Social Sciences, defined E-commerce as data (information) exchange activities of the buyer and the seller carried out via modern information technology. In The E-commerce of the 21st Century compiled by Wangjun in Chinese Journal of Computers, E-commerce refers to a whole set of dynamic technology, application software and business process that links enterprises, customers and the society. Prof. Wangjian from the University of International Business and Economics stated in his E-commerce lecture that E-commerce refers to an electronic process of realizing a whole transaction process by electronic information technology, network interconnection technology and modern communication technology, where the transaction parties only need to contact with each other through internet instead of transmitting paper documents or files (Chen D N, Jeng B, Lee W P., 2008). By taking advantage of Agent technology in artificial intelligence, this paper made a classification based on the business properties of different business activity parties, established search engine BP neural network model on related E-commerce platform, and improved the commercial negotiation function of the platform.

2. Related Theory and Technology

2.1. Definition of Agent

In multi-Agent system, Agent is an independent entity, which continuously interacts with the environment. In the MAS environment, however, there are other processes with other Agents. In another word, Agent is an entity with a state consisted of mental components as belief, ability, selection and intention, etc. Agent can be homogeneous or heterogeneous in the system. The research of multi-Agent is to manage the intelligent behaviors of a group of independent intelligent Agents, coordinating their knowledge, ability, selection, intentions and plans to take joint measures or solve problems together. There are both cooperation and competition between the Agents. A common feature of distributed artificial intelligence and multi-Agent system is the distributed entity behavior. The multi-Agent system is designed from the bottom up. Because in principle, distributed and independent Agents are first defined, then study how to accomplish the task solving of a single or several entities. Multi-Agent system can deal with both single target and different targets. Multi-Agent system mainly studies how the logically or physically separated Agents cooperate and calculate simultaneously to solve problems (Carvalho, A. A., Araújo, I., & Fonseca, A., 2015). The main purpose is to analyze and design large and complex cooperative intelligent system, such as large information system and intelligent robot, etc. Recently multi-Agent system is studied a lot. MultiAgent system attempts to use Agent to simulate human's rational behaviors, which is mainly applied in the fields of simulation on reality and society, robots and intelligent machinery, etc. Agent itself should be possessed with properties as independence, interactivity with environment, collaboration, communication, longevity, selfadaptability and instantaneity, etc. (Sandholm T., 2002)

2.2. Classification of E-commerce

Based on the business properties of different business activity parties, E-commerce can be divided into following categories:

1. B2B (Business-to-Business) E-commerce:

B2B E-commerce refers to the E-commerce activities between enterprises. For example, industrial and commercial enterprises use computer network to purchase from their suppliers or make payments. This kind of E-commerce, especially business activities conducted by Electronic Data Interexchange (EDI) through private or Value-addednetwork (VAN), has been existed for years.

The EDI technology that supports B2B refers to the exchange of unified structure and standard information conducted by entities through computer network. As the technology supports a direct exchange of information between computer systems, it can reduce or even eliminate artificial intervention and information input to the maximum. At present, EDI is successfully applied in the field.

Although the implementation of EDI is facing many difficulties, B2B E-commerce is still the main trend of E-commerce as seen from the perspective of future development.

2. B2C (Business-to-Consumer) E-commerce:

B2C E-commerce refers to the E-commerce activities between enterprises and consumers. This kind of E-commerce mainly uses the online selling activities through international internet. In recent years, B2C E-commerce has sprung up with the development of international internet. For example, there are a lot of supermarkets on the internet, where everything is available, including commodities, drinks, computers, and even automobiles, etc.

The reason for the rapid development of B2C E-commerce is because the development of international internet has created a new transaction platform for enterprises and consumers.

From the technical perspective, there is no need for a unique standard of data transmitting for enterprises and consumers. Online retail or payments only involves credit card or other electronic currencies. In addition, the browsing function and multi-media provided on the international internet make it easier for consumers to find what they need and know more about the commodities. Hence, B2C E-commerce has the minimum obstruction and large potential. Obviously, this kind of E-commerce will keep on developing and is one of the motives in promoting other types of E-commerce activities.

3. C2C (Consumer -to-Consumer) E-commerce:

C2C E-commerce is the E-commerce activities between consumers, which mainly takes advantage of the online selling activities through international internet.

4. B2A (Business-to-Administrations) E-commerce:

B2A E-commerce refers to the E-commerce activities between enterprises and authorities.

5. C2A (Consumer-to-Administrations) E-commerce:

C2A E-commerce refers to the E-commerce activities between authorities and individuals. This kind of E-commerce has not been really formed (Kartiwi M, MacGregor R C., 2007).

2.3. Genetic Algorithm (GA)

Genetic Algorithm (GA) is proposed by Prof. Holland from the University of Michigan in 1975. It is originated from Darwin's theory of evolution, Weizmann's theory of natural selection and Mendel's population genetic theory. GA is a self-organizing and adaptive artificial intelligent technology that simulates the nature evolution procedures and mechanism to solve extreme problems. The basic idea of GA is to form an algorithm of process searching optimal solution based on the simulation of natural genetic extreme value and biological evolutionism, which has a firm biological foundation. GA is a collateral global search algorithm which adopts probabilistic random seeking strategy which can automatically obtain related information in the searching period and adaptively adjust searching directions with strong robustness. Nowadays GA has been widely used in the fields of optimal control, signal processing, combinatorial optimization and artificial life, etc. (Nguyen G, Dang T T, Hluchy L., 2007)

In the simulation of natural genetic recombination and evolution, GA will first encode the problems to be solved. Every unit in the code is called the "Gene"; a group of genes can form a "Chromosome", known as an individual; several chromosomes will be processed repeatedly by operators as selection, crossover and mutation, until an optimal or sub-optimal solution is obtained. The basic operation process of using GA to solve problems is shown as Figure 1.

1. Code: code is the primary issue to be solved in GA, which can turn the practical solution into genetic structure data;

2. Generate initial species: Randomly generate initial species consisted of K genes;

3. Individual evaluation: Evaluation the advantages and disadvantages of each chromosome with adaptive functions;

4. Selection computation: The purpose of selection is to directly pass down good individuals to the next generation based on the adaptive evaluation of individuals.

5. Crossover computation: Part of two parent individual structures are replaced and recombined to generate a new individual. Crossover is actually the operation of information exchange between parent individuals;

6. Variation computation: Change the genes of some individual chromosomes in the specie to increase the variety of the specie and improve the local search ability of the algorithm;

7. Determination of termination conditions: The algorithm terminates when the adaptivity of the optimal individual reaches the given threshold value or the adaptivity of the optimal individual and the specie stops rising (Li L, Horrocks I., 2004).

2.4. Negotiation Model based on GA

Whether the negotiation of E-commerce reaches an agreement or not would decide the success or failure of the whole business activity. A good negotiation model should satisfy the following requirements:

1. Can describe the negotiation problem effectively and comprehensively;

2. Can describe the human factors in the negotiation;

3. The required computing resource is limited during the negotiation process.

Based on the above requirements, this paper designed a bilateral multi-topic negotiation model based on GA.

As the negotiation model can sense the specific environment and represents the user to accomplish a series of goals, this paper made the following basic hypotheses in the negotiation model:

1. Agent is selfish. It pursues the maximum benefit of their side;

2. Agent is not complete. It has no idea of the preferences of the opposite.

3. Agent has a limited rationality. The strategies it adopted are usually the most optimal strategies that are beneficial to them;

4. There is no cheating behaviors in the negotiation. Both of the negotiation parties are willing to reach an agreement;

5. Time is precious for the negotiation parties, which means they are making a limited negotiation;

6. Failure of negotiation is the worst result for the two parties.

The negotiation model in the paper can be defined by the following multi-component system, as shown in Formula (1).

... (1)

Ag represents the Agent sets in the negotiation, Agi ∈ Ag. Here we assume i∈(b,s), where Agb stands for customer negotiation Agent, Agb represents the enterprise negotiation Agent.

Q stands for negotiation project sets, Q= {Q1,L,Qj,LQJ, Qj refers to the jth negotiation project (the jth gene), project number J ≥ 2 . The negotiation project sets in this paper include commodity price, delivery time and warranty.

X is the value of the negotiation project,

... (2)

Where x j stands for the proposition value of property Qj in the negotiation Agent. All the proposition value should be included in the given range of xj∈ ??xmjin,xmj ax ?? , which is also the expected range of each negotiation Agent.

W stands for the project weight set,

... (3)

Wji is the weight of Agi to Q j, the value of which is designated by the users of the two sides. For example W1i> W2i, means Agi prefers Q1 for the project Q1 and Q2.

In this paper, assume the weight sum of the user to all the negotiation projects is 1,

...(4)

V: evaluation function, representing the evaluation of Agent on some value in the negotiation project set. The evaluation function can be linear or non-linear. For simplicity sake, this paper takes monotone linear function as the evaluation function. The monotone non-decreasing function is shown as Formula (5).

...(5)

Monotone increasing function is shown as Formula (6)

...(6)

During the negotiation, the merchant and consumer Agent will select proper evaluation function for each project according to their different interest orientations of the project. For the same project, their evaluation function is at the opposite.

U stands for utility function, which is the evaluation of user negotiation Agent on the value of some negotiation project set. The calculation method is shown as Formula (7).

...(7)

T: Time limit of negotiation

A: Motion set of negotiation

A= {call,request,propose,agree,reject (8)

Where, call means to send the negotiation request; request means to respond the negotiation request; propose means to send proposal; agree means to agree the proposal; reject means to refuse the proposal (Sierra C., 2004).

2.5. Establishment of BP Neural Network Model in Search Engine

BP (Back Propagation) neural network is a learning algorithm of neural network. The full name is the artificial neural network based on error back propagation algorithm. The single-hidden layer feed forward neural network of topological structure is generally known as three-layer feed forward network or three-network perceptron, including the input layer, middle layer (or hidden layer) and output layer. The characteristics of the network is that the neurons of each layer only fully connect with the neurons of the neighboring layer, and there is no feedback connection between the neurons of each layer, which form the feed forward neural network system with hierarchical structure. The single-hidden layer feed forward neural network can only solve linear and separable problems. Only multi-hidden layer neural network can solve the non-linear problems.

1. Manually input given keywords in the search engine "Hui Sou", download training sample articles through internet, and extract Chinese words. In the previous experiments, 10000 articles of each webpage of employment, company information and renting information were downloaded respectively. Among all the information, employment information and company information has the highest similarity, which can be tested to decide the training ability of network.

2. Use the interface provided by the open-source program of "Huge word segmentation" to conduct word segmentation of the downloaded Chinese information and extract the keywords with high occurrence rates.

3. Establish neural network: According to the original data characteristic and anticipated classifying effect, neural network tends to adopt input layer, single hidden layer, output layer and 3-layer topological structure. Obtaining of input neuron: Input neuron refers to the weights of the preliminary extracted keywords in the sample article (binarization processed with keyword "1", otherwise "0"). Selection of activation function: sigmoid activation function (hidden layer and output neuron):

...(9)

3. Agent Technology in E-commerce

3.1. Weak Definition of Agent

At present, there is no confirmed and unified definition of Agent in academia. Researchers gave different definitions on "Agent" according to their different research contents and targets. Wooldridg and Jenning summarized the definitions of "Agent" and provided the definition that is basically accepted by the academia. They gave "Agent" two definitions as weak definition and strong definition according to the usage of "Agent".

The weak definition of "Agent" refers to the computer software or hardware system with properties shown in Figure 2.

Autonomy: it is the basic property of Agent, which can control its autologous behaviors and internal state. Without human's direct interference, Agent can also activate a continuous motion. Its behavior is active and spontaneous; Agent has its own goal and intention; According to the requirement of target and environment, Agent should make plans for its short-term behaviors.

Social ability: Agent Communication Language (ACL) can be used to interact with other Agents. As a basic property of Agent, social ability is also known as communication, which means the information exchange between the Agents. Furthermore, Agent can proceed with "conversation". Tasks undertaking, multi-Agent cooperation and negotiation all take the social ability or communication as the basis.

Reactivity: it means the awareness and influence to the environment. "Agent" has the ability of exploring the environment condition and responding the environment. No matter Agent lives in the real world (such as robot, service Agent on Internet) or virtual world (such the Agent in virtual mall), all the Agents should be able to perceive the environment and change the environment through their behaviors. The object that is unable to influence the environment cannot be regarded as an Agent.

Proactivity: "Agent" can take proactive measures to prove its goal-oriented property. Temporal continuity: also known as longevity. Traditional programs are activated when needed and terminated when not needed or computation completed. Different from the programs, Agent can continuously work in a "pretty long" period. Although proactivity is not a must property, it is considered as an important property of Agent (Klose M, Lechner U., 1999).

3.2. Strong Definition of Agent

Except for the properties of weak definition, strong Agent also has following one or several properties:

Mobility: Agent has the ability of moving on computer network.

Rationality: the action of Agent should help to reach its goal rather than prevent from its goal. At least in the allowable range of belief, conflict targets are not accepted.

Adaptability: Agent should have the ability of self-adjustment and be able to adapt with the working methods or ways of other users.

Collaboration: Agent doesn't live alone but with other Agents. The effective and efficient cooperation of Agents can greatly improve the performance of the whole system. Besides, Agent cannot accept and execute any order. It should note that people would make mistakes and ignore some important information or provide uncertain information. Agent can detect the case and solve the problem through the user model established by the system or even refuse to execute the task (Terpsidis I, Moukas A, Pergioudakis B., 1997).

3.3. Problems for Agent

Common Agents have to solve the problems as cooperative mode, workflow management, media space share, heterogeneous resource integration, collaborative system security, collaborative application development environment, and virtual cooperation environment, etc. E-commerce Agents have their own technical difficulties as shown below except for the above problems:

Information discovery problem. There are a lot of distributed networks on Internet. It has been a hard-to-solve problem to discover the most appropriate information from these networks. At present, the professional information of a commodity is usually distributed everywhere on the Internet. Some company or specialize organization created the website and provides customers with the information. With the increase of online information, it becomes even harder to search and integrate the important information. Both of the buyer and seller have the same problem. Now "keywords searching" is the main solution to information discovery. However, the range of this method is limited. In addition, as the method cannot realize the related search of synonyms, "keywords searching" technology is not appropriate to search all the data on the Internet in a specified domain range. Hence search engines as "YAHOO" (www.yahoo.com) are needed for Agent to obtain all the related information of certain information. XML(Extensible Markup Language) also provided a solution to the problem of information discovery.

Communications problem: in order to obtain the professional and detained information of a certain commodity, the buyer Agent needs to contact with the seller Agent, which may lead the following three problems:

1. As shown in Figure 3, the buyer and seller should adopt the same transport protocol to transport demands and results. Protocols as TCP/IP can solve these problems by unified and convenient transportation of demands and results.

2. The next problem is to design a public transport language. Each party in the E-commerce should understand the language. At present, HTML is a popular language on Internet. HTML format text allows the browser to explain the webpage on WWW. However, as the main purpose of HTML design is to accurately display information rather than extract information, the information that Agent needed are usually embedded in a specific HTML text. It has always been the goal of KQML (Knowledge Query and Manipulation Language) to understand how to obtain useful information from HTML and transport them. KQML is a transport language of Agent. Similar Agent communication languages include FICA ACL (Agent Communication Language) and MIL (Market Interaction Language) which are still at the developing stage. MIL is a language specially designed for the communication of E-commerce Agents.

3. There is only one communication problem left after solving the transport protocol and transport language problems. The buyer and seller need to negotiate on the transport content, which requires the formulation of the content expression methods in the public transport language (Lin F, Huang S, Lin S., 2002).

3.4. Design and Realization of E-commerce System based on Mobile Agent

3.4.1. Systematic Structure of E-commerce Agent

This part summarized three systematic structures of E-commerce Agent:

Purchase Agent with simple ability can intellectually guide the users to buy certain commodities. All the operations of this type of Agent should observe the following two principles:

1. Search-supporting principle. The target of Agent with simple purchasing ability is to provide certain commodities for users. To realize this, service Agent should be equipped with the interface provided by the merchants. As soon as the users define the searching conditions, Agent with simple purchasing ability will query the backstage commodities base of each merchant through the interfaces, summary the obtained results and transport back to users in a standard expression form. The Agent with simple purchasing ability can only search the particular commodities of a certain domain. Users can only query the commodities provided by the merchants specified in the domain.

2. Price comparison principle. According to users' requirements, multiple merchants may return the commodities that satisfy the purchasing conditions of the users. When displaying to users, the prices, quality, service providers and additional charges (such as postal charge) will be briefly given for users' selections, which can save their time and facilitate their selections. Based on these two principles, Agent can accomplish the selection tasks of commodities (not real purchasing tasks) for users. Then, users can contact with the specified merchants and accomplish the real transactions. Based on the above operation process, this paper presented the systematic structure of the Agent with simple purchasing ability as following.

This systematic structure contains four main hierarchies:

1. Query interface: In the query interface, the search conditions of the required commodities will be defined and the returned query results will be described.

2. Sub-Agent: Sub-Agent is used to execute the query task and report the results to the administrator. The sub-Agent will contact with a certain merchant, search the backstage data base of the merchant and seek the commodities that satisfy the requirements of users. After obtaining the results, sub-Agent would forward the commodity information to the administrator. All the operations are conducted simultaneously. One sub-Agent contacts with one merchant, which greatly reduced the time of searching.

3. Administrator: The administrator is responsible for coordinating the actions of each sub-Agent and the organization, combination and ranking of the result information of each sub-Agent. He has all the information of the registered merchants and the sub-Agent at the corresponding interfaces. When receiving a specific request of the user, the administrator should transfer the request in time. When receiving specific commodity description from the merchant, the administrator should classify the results according to certain intelligence and transfer to the query interference for user's browsing.

4. Merchant: Merchant provides the commodity data base for the purchasing Agent with simple ability. The sub-Agent query will do the search on the data base. The purchasing Agent with simple ability has no direct effect on the merchant (Lee K Y, Yun J S, Jo G S., 2003).

3.4.2. General Design of System

The system is designed on the basis of MVC, which mainly include four hierarchies: expression hierarchy, Agent hierarchy, application server and data base server, as shown Figure 5.

1. Expression hierarchy

The user interface adopts the browser method to facilitate the user. As different users have different permission, this system has three different user interferences based on the transaction process of E-commerce, including merchant interface, administrator interface and buyer interface. Hence, different service interfaces will be presented to users with different permissions.

2. Agent hierarchy

Agent hierarchy is an indispensible part of realizing the mobile Agent E-commerce system. It mainly includes buyer Agent sub-system, merchant Agent sub-system, logistics Agent sub-system, payment Agent and intermediate Agent, etc. When transaction occurs, each Agent system would cooperate and accomplish the transaction rapidly and conveniently.

3. Web server

Web server is also known as the intermediate hierarchy. It mainly conducts the logic processing and maps to related data base hierarchy. The main function module of the system includes user administration, transaction administration and after-sales processing, etc.

4. Data base server

As an important part of the system, data base mainly stores various information needed in the transaction process. The system data base mainly consists of buyer data base, merchant data base, logistics data base and intermediate data base, etc. (Puliafito A, Tomarchio O, Vita L., 2000)

4. Experimental Data and Analysis

The target of the experiment is to download the E-commerce information on the internet through the search engine model "Hui Sou" designed in this paper, use neural network algorithm to automatically classify the above information, compare the number of iterations and classification accuracy of the classification algorithms by adjusting parameters as the number of neurons and training samples, so that to testify the feasibility of neural network algorithm in the intellectual search engine.

The number of input neurons f BP neuron network is 45, mainly includes the keywords with higher appearance selected by mapping keywords from three types of information (each type 15 neurons). The result of the selection is shown as below:

When the sample number is 1000, the test times is 200. When the training sample number is 100, the experimental data is shown as Table 1:

According to the experimental result, when the number of BP neural network input neurons is 45, all the iteration number of the classification algorithm is less than 10000, and the accuracy is higher than 0.9, as shown in Figure 6 and Figure 7. The established neural network model has good generalization ability, which can achieve the expected classification effect and apply to intellectual searching engine model. Besides, according to the comparison, the number of neural network input neurons has limited effect on the whole classification result, and the iteration number and accuracy are basically in the same level range (45 input nodes and the iteration number is relatively low). In this case, the selection difficulty of BP neural network input neurons can be avoided.

Due to the network property of "Work after learning" of BP neural network and high work efficiency, although there are much iterations during the network training, the training process can be offline operated, which will not affect the visit speed of internet. The experiment also proved that the classification algorithm can satisfy the speed requirement of search engine (Su C J., 2008).

5. Conclusion

With the new trend of times, E-commerce has been more and more popular, and gradually leads the economic development of modern society. Due to the rapid information transfer, low transaction cost, high-efficient circulation, and large market scale of E-commerce, various enterprises set foot in E-commerce, which promoted the continuous development of our economy. This paper gave a definition on Agent technology, including weak definition and strong definition, classified E-commerce according to different business properties, introduced GA, proposed the negotiation model based on GA and established the general design based on mobile Agent E-commerce system on the basis of the above theory. The mobile Agent E-commerce system mainly consists of four hierarchies as expression hierarchy, Agent hierarchy, application server and data base server. In addition, the paper also applied BP neural network model into search engine and testified through experimental analysis that the classification algorithm has high work efficiency and can satisfy the speed requirement of search engine. In conclusion, this paper proposed a comparatively perfect E-commerce platform based on artificial intelligence, which can effectively realize intellectualized E-commerce and solve the problems that may appear in the using and development of E-commerce system.

Acknowledgment

This work was supported by the Research on the middleware technology of agricultural products in the Yellow River Delta; Binzhou Science and Technology Development Plans (No. 2011ZC0402).

References

References

Carvalho, A. A., Araújo, I., & Fonseca, A. (2015). Das Preferências de Jogo à Criação do Mobile Game Konnecting: um estudo no ensino superior. RISTI - Revista Ibérica de Sistemas e Tecnologias de Informação, 2015(16), 30-45.

Chen D N, Jeng B, Lee W P. (2008). An agent-based model for consumer-to-business electronic commerce. Expert Systems with Applications, 34(1), 469-481.

Glushko R J, Tenenbaum J M, Meltzer B. (1999). An XML framework for agent-based E-commerce. Communications of the ACM, 42(3), 106-ff.

He M, Jennings N R, Leung H F. (2003). On agent-mediated electronic commerce. Knowledge and Data Engineering, IEEE Transactions on, 15(4), 985-1003.

Kartiwi M, MacGregor R C. (2007). Electronic commerce adoption barriers in small to medium-sized enterprises (SMEs) in developed and developing countries, A cross-country comparison. Journal of Electronic Commerce in Organizations, 5(3), 35-51.

Klose M, Lechner U. (1999). Design of business media-an integrated model of electronic commerce. AMCIS 1999 Proceedings, 193-200.

Lee K Y, Yun J S, Jo G S. (2003). Mocaas, auction agent system using a collaborative mobile agent in electronic commerce. Expert systems with applications, 24(2), 183-187.

Li L, Horrocks I. (2004). A software framework for matchmaking based on semantic web technology. International Journal of Electronic Commerce, 8(4), 39-60.

Lin F, Huang S, Lin S. (2002). Effects of information sharing on supply chain performance in electronic commerce. Engineering Management, IEEE Transactions on, 49(3), 258-268.

Nguyen G, Dang T T, Hluchy L. (2007). Agent platform evaluation and comparison. Rapport technique, Institute of Informatics, Bratislava, Slovakia.43-49.

Puliafito A, Tomarchio O, Vita L. (2000). MAP, design and implementation of a mobile agents' platform. Journal of Systems Architecture, 46(2), 145-162.

Sandholm T. (2002). eMediator, A next generation electronic commerce server. Computational Intelligence, 18(4), 656-676.

Schubert P, Ginsburg M. (2000). Virtual communities of transaction, The role of personalization in electronic commerce. Electronic Markets, 10(1), 45-55.

Sierra C. (2004). Agent-mediated electronic commerce. Autonomous Agents and Multi-Agent Systems, 9(3), 285-301.

Su C J. (2008). Mobile multi-agent based, distributed information platform (MADIP) for wide-area e-health monitoring. Computers in Industry, 59(1), 55-68.

Terpsidis I, Moukas A, Pergioudakis B. (1997). The potential of electronic commerce in re-engineering consumer-retail relationships through intelligent agents. Advances in Information Technologies, the business challenge, 29, 64-70.

Tsvetovatyy M, Gini M, Mobasher B. (1997). Magma an agent based virtual market for electronic commerce. Applied Artificial Intelligence, 11(6), 501-523.

Wang W, Benbasat I. (2007). Recommendation agents for electronic commerce, Effects of explanation facilities on trusting beliefs. Journal of Management Information Systems, 23(4), 217-246.