The role of intelligent agents and data mining in electronic partnership management

Expert Systems with Applications 39 (2012) 13277–13288

Contents lists available at SciVerse ScienceDirect

Expert Systems with Applications

journal homepage: www.elsevier .com/locate /eswa

The role of intelligent agents and data mining in electronicpartnership management

Merrill Warkentin a, Vijayan Sugumaran b,c,⇑, Robert Sainsbury d

a Department of Management and Information Systems, Mississippi State University, P.O. Box 9581, Mississippi State, MS 39762-9581, USAb School of Business Administration, Oakland University, Rochester, MI 48309-4401, USAc Department of Global Service Management, Sogang Business School, Sogang University, Seoul 121-742, Republic of Koread Gravity Jack, Spokanne, Washington, USA

a r t i c l e i n f o

Keywords:Intelligent agentsElectronic partnershipSupply chainData miningXML

0957-4174/$ - see front matter � 2012 Elsevier Ltd. Ahttp://dx.doi.org/10.1016/j.eswa.2012.05.074

⇑ Corresponding author at: School of Business Admsity, Rochester, MI 48309-4401, USA. Tel.: +1 248 370

E-mail addresses: [email protected] (M.land.edu (V. Sugumaran), [email protected] (R. S

a b s t r a c t

The marketspaces of the ‘‘New Economy’’ and the eServices revolution have enabled the formation of newtypes of partnerships which are electronically mediated. Web-based electronic commerce has alsobrought a tremendous increase in the volume of data that can be mined for valuable managerial knowl-edge. The data mining procedures used in this process can be enhanced by employing intelligent agents.This paper describes emerging electronic partnerships between players in developing electronic market-spaces and identifies typical data flows between such players, with an analysis of the potential role ofdata mining and intelligent agent technology. By identifying the complex nature of information flowsbetween the vast numbers of economic entities, we identify opportunities for applying data mining tech-niques that can lead to knowledge discovery. In particular, we show how a Generic Agent-based dataMining Architecture (GAMA) can be customized to support managerial decision-making and problemsolving in a networked economy. A prototype implementation of GAMA is presented, along with a dem-onstration of the some of the capabilities of the system. Finally, we explore the role of agents in promot-ing and maintaining strong automated relationships between various strategic partners.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Millions of individuals surf the Web every day and interact withelectronic commerce Websites around the world. While many sitescapture user activity, most do not capture all interactions with‘‘etail’’ (electronic retail) consumers, suppliers, and partners, andthey do not maximize the potential uses for such data (Liu, Cao,& He, 2011; Willow, 2005). According to Forrester Research, only18% of the companies it surveyed use their Web data for marketingpurposes, and only 16% use it for customer support. The Forresterstudy also indicates that 72% of the companies that collect Webdata admit that they do not analyze this data or use it in any appli-cation. However, organizations are beginning to realize the value ofthis Web data and are allocating vast resources for creating thenecessary infrastructure to analyze this data, which would enablethem to learn more about their customers and gain a competitiveadvantage (Rajan & Saravanan, 2008; Yarom, Rosenschein, &Goldman, 2003; Zuo & Hua, 2012).

ll rights reserved.

inistration, Oakland Univer-2831; fax: +1 248 370 4275.

Warkentin), sugumara@oak-ainsbury).

The sheer volume of data generated by the activities of visitorsto a company’s site (their ‘‘digital footprint’’) poses various prob-lems in the storage, management, and analysis of this data as wellas new opportunities. Some companies rush to set up theirelectronic storefronts, focusing more on transaction processing,online inventory, shopping carts, and ad banners, without givingsufficient consideration to the data management issues (Gomes &Canuto, 2006; Holmes, Tweedale, & Jain, 2012). In order to getthe most mileage out of this data, each company must decide:(a) what data to collect and how to organize it; (b) what kind ofanalysis to perform on the data; (c) how frequently to perform dataanalysis; and (d) how to validate and integrate the results intodecision making and planning.

As data warehousing and data mining technologies mature, anincreasing number of organizations are employing these technolo-gies in their problem solving and managerial decision making inthe business to consumer context (Rao, 2010). Through data min-ing, a company can synthesize consumer Website patterns intomeaningful information, enabling it to understand and engage cus-tomers and prospects over the Internet (Chaimontree, Atkinson, &Coenen, 2012; Nassiri, 2009). The mining of Web-based data andthe implementation of the business intelligence it represents isthe key to creating a lasting relationship with online customers

1 Transmission Communication Protocol / Internet Protocol, HyperText TransferProtocol, HyperText Markup Language, and eXtensible Markup Language.

2 US-based Fortune 1000 companies spend approximately half their budgets onvarious categories of procurement, ranging from major acquisitions (such as rawmaterials and sub-assemblies purchased by heavy manufacturers) to routine officesupplies.

13278 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

and establishing a successful online storefront. But in the future,mining the Web service interactions between company Websiteswill also enable them to create lasting relationships with their stra-tegic partners (Jain, 2012; Marik & McFarlane, 2005).

There are several commercial software products available toanalyze Web traffic data such as NetTracker, WebTrends, NetIntel-lect, HitList, and SurfReport. However, these products are limitedto analyzing server activity based on the data stored in log files.By unifying the log data with personal information supplied byvendors such as Equifax, Experian, TransUnion, MetroMail, andothers, one can develop a more complete customer profile. Thisintegrated information can then be mined to gain insight intowho is buying what products, what products are the most popular,buying patterns, and so forth (Gao, Yan, & Dang, 2012; Jayabrabu,Saravanan, & Vivekanandan, 2012; Kehagias & Mitkas, 2007). Thedynamic nature of the online environment dictates that this anal-ysis should be performed promptly to enable companies to quicklyrespond to changes in customers’ buying behavior.

There are myriad data mining tools available in the market thatemploy a variety of data mining algorithms and techniques. For anovice user, it is often difficult to determine which tools ortechniques are appropriate for a particular data analysis or datamining scenario. Companies are beginning to employ ‘‘intelligentagents’’ (Chan, Fan, Prodromidis, & Stolfo, 1999; Gannon & Bragger,1998; Gorodetsky, Karsaev, Samoylov, & Serebryakov, 2008;Grimes, 1998; Sugumaran & Bose, 1999) to reduce some of thiscognitive load. These agents can automate some of the mundaneactivities such as data cleansing and data transformation andcan help the user in the selection of appropriate tools and datamining methods (Li & Li, 2011; Lee & Liu, 2004; Moemeng, Zhu,Cao, & Jiahang, 2010). Typically, intelligent agents act on behalfof the human user in problem solving activities and decisionmaking.

The objective of this research is to: (a) study the informationflow between various entities in different electronic markets; (b)investigate how data warehousing and data mining techniquescan be applied for discovering new relationships and nuggets ofknowledge that could be incorporated into managerial decisionmaking; and (c) develop a generic architecture for an intelligent-agent based data mining environment; and (d) apply this architec-ture to various eCommerce marketspaces to help the user validateand interpret the results, thereby enabling the discovery of valu-able knowledge.

The remainder of this paper is organized as follows. The nextsection (‘‘Emerging eCommerce Marketspaces and Data Flows’’)presents a vision of relationships between strategic partners facil-itated by the implementation of interoperable automated pro-cesses. The following section (‘‘Relationship Management in theNew Economy’’) focuses on collaborative commerce (cCommerce)activities such as outsourcing and establishing strategic partner-ships through the use of intelligent agents. The next section(‘‘Emerging Technologies’’) provides an overview of emerging tech-nologies such as data warehousing, data mining, and intelligentagents, and shows some examples of their use on the Web. Thefollowing section (‘‘Generic Architecture for Agent-Based DataMining’’) proposes an architecture for a generic agent-based datamining environment. This Generic Agent-Based Data MiningArchitecture (GAMA) can be customized to support managerialdecision-making and problem solving for a particular application.The penultimate section (‘‘Agent-based Data Mining Applicationsin eCommerce’’) provides a detailed discussion of the applicationof intelligent agent-based data mining in different electronic mar-ketspaces such as etailing and B2B exchanges. The final section(‘‘Managerial Implications and Future Directions’’) concludes thepaper by discussing the issues in agent-based data mining andtheir managerial implications.

2. Emerging eCommerce marketspaces and data flows

The term ‘‘electronic commerce’’ has been used to describe adisparate variety of business activities. The focus in this paper willbe on relationships among various economic entities (individualsand Websites) which are supported by the standard communica-tion and data representation protocols of TCP/IP, HTTP, HTML,and XML.1 Many such relationships are built on data flows in thebackground of the business-to-consumer (B2C) channel or betweentwo business computers (business-to-business or B2B) which aretransparent to the human user (Zhang, Wang, & Shen, 2012).

The adoption of electronic commerce technologies has enabledentirely new economic market models that connect many firms toconsumers (etailing or B2C eCommerce) and to other businesses(B2B eCommerce). Associated with these emerging markets arenew strategic business models, frequently built upon strong strate-gic alliances with firms that provide critical services in the valuechain (or value web).

Business-to-Consumer eCommerce (or etailing) has evolvedfrom an obscure commercial niche to a widely-accepted marketingchannel for selling countless retail items. A number of factors dis-tinguish etailing from traditional markets, but the most importantone is its implementation of electronic customer relationship man-agement (CRM). Etailers can effectively analyze buyers’ clickstreamdata and purchase data in order to offer personalized services andproduct offerings (Tuzhilin, 2012). An online bookseller, for exam-ple, may display an opening screen that features items similar tothose a specific customer has purchased or researched in the past,and it may send email reminders when new items arrive that mayinterest the buyer. The online seller may also enable flexible order-ing processes, online shipment tracking, and quick checkouts.

Business-to-business eCommerce is evolving from a proprietaryEDI based, closed, expensive, and non-scalable system to an open,inexpensive, and scaleable system where multiple suppliers andbuyers are connected to each other by Web-based exchanges. Be-cause procurement activities comprise such a significant portionof many companies’ budgets,2 the greater selection and transpar-ency (price, availability, supplier, product) that the B2B systembrings has been largely welcomed by the capital markets. Vortals(industry-specific vertically integrated portals) and e-procurementrepresent the majority of B2B activity.

In a later section, we analyze the role of data mining and intel-ligent agents in B2B eCommerce where multitudes of buyers andsuppliers interact to share information regarding products andconduct transactions using a variety of market mechanisms likecontract procurement, auctions, reverse auctions, and bid-ask spotmarkets. The presence of multiple players and complex informa-tion flows between them creates interesting possibilities forknowledge discovery using data mining. Agent-based data miningcan facilitate the establishment of automated relationship man-agement by identifying trends in purchase activity, profiling busi-ness partner activities, and offering support for various onlineexchange activities (Liao, Yang, & Geng, 2011).

3. Relationship management in the new economy

In today’s rapidly changing eBusiness environment, it is imper-ative that organizations have flexible organizational structures thatcan accommodate partnerships with a variety of external

M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288 13279

companies in a dynamic and agile ‘‘virtual organization’’ (Khalil &Shikuta, 2012; Mowshowitz, 1997). The goal is the development ofthe truly ‘‘agile corporation’’ in which a firm engages in continuousreconfiguration as it makes frequent choices about partnershipsthat best fit the demands of the moment. Many firms form alli-ances and partnerships in one business area with firms that arealso competitors in other business areas. The Web enables greatercollaboration and virtual partnership, and will enable much morein the future as standards are developed (such ebXML, RosettaNet,UDDI (Universal Description, Discovery, and Integration), and WebServices Description Language (WSDL). With real-time digitalinteraction between various front-office and back-office functions,two or more firms can quickly form a business-to-business collab-orative network with little requirement for exhaustive configura-tion and design steps that would have been necessary in the past(Xu, Tong, & Tan, 2011). With industry standard applications,methods, and data representation schemes, two or more firmscan ‘‘plug into each other’’ with little or no human effort. This pro-cess has been termed ‘‘collaborative commerce’’ or cCommerce.Carr (2000) suggests that, rather than being primarily a tool for dis-intermediation, the Web is proving to be a new channel for inter-mediation and reintermediation. He suggests that we are enteringan age of hypermediation, in which all firms extensively use theservices of other firms provided over the Web (Web services) ina great pool of outsourcing opportunities. Fig. 1 demonstrates thecomplexity of relationships that can exist even for a simple B2Cconsumer portal site (without even considering many of the pro-cesses that are required).

Engaging in cCommerce activities across dynamic digital net-works of partners requires a sophisticated new approach to man-agement (Warkentin, 2002). Rather than managing the processesdirectly, many modern managers must establish and monitor rela-tionships with outsourcing entities that serve the overall missionand objectives of their firms. A firm may engage in ‘‘selective out-sourcing’’ to leverage a provider’s unique competence for a specificbusiness function, or it may practice extensive outsourcing to asingle end-to-end service provider or to multiple entities. In eithercase, it is imperative that the firm establish a clear understandingwith all of its outsourcers regarding objectives, outcomes,

Content Syndicator

(acquires and

redirects content)

Adserver Ne(broker of a

Advertisers

Content Sources

B2C Web (etailer, por

Content DelOptimize

B2C Custo

Internal B2B Services

- hosting, maintenance - payment processing - sales tax calculation - encryption, dig wallet - data mining services

Fig. 1. B2B services infrastructure (So

measures, revenue agreements, information sharing arrangements,and other factors (Jiang & Zhao, 2012). Rather than initiating anadversarial position, both parties must ensure a true partnershipof trust and common interest. By linking part of a vendor’s com-pensation to the tangible benefits it delivers, the partner is incent-ed to contribute to the firm’s goals. To determine the success of apartnership relationship, there are third-party services that willmonitor and log various transactional metrics in an online ex-change. For example, third parties can independently verify thevolume of downloads from a site, the number of banner ads dis-played, or the clickthroughs from an affiliate’s Website.

Cunningham (2001) offers a framework for determining thepartnership arrangements that are appropriate for various circum-stances. He identifies inside partnerships, ‘‘nearside’’ partnerships,and network partnerships that play various roles in B2B eCom-merce. Further, he suggests that newer, simplified agreementsand processes have made it easier to enter into partnerships with-out time-consuming negotiation.

When evaluating selective process outsourcing versus totalend-to-end service providers, managers should carefully considerthe trade-offs. Rather than managing multiple relationships, a sin-gle provider may provide a streamlined approach to partnershipmanagement. However, if some of those services are not the ‘‘bestof class,’’ a firm may prefer to use separate vendors for each out-sourced process. This latter approach may ensure higher levels ofservice, creating a more successful overall eCommerce solution,but at the cost of greater partnership coordination. The firm mayhave to operate multiple protocols and systems, which is organiza-tionally confusing and expensive.

Modern Business-to-Exchange (B2X) networks, which utilizeXML-based technologies such as WSDL, UDDI, and SOAP, enable afirm’s systems to ‘‘plug in to’’ web services and utilize them on apay-for-use basis without complicated negotiation and systemconfiguration management (Hong, 2010). Fig. 2 shows a B2X Hubwhich could connect an enterprise to many exchanges, all usingstandard communications and applications protocols. The abilityto ‘‘snap together’’ various applications and functionalities wouldprovide a new venture the ability to take a nascent idea and forma complete transactional Web-based virtual corporation in a short

twork ds) Affiliate

Services Provider (and other sources of traffic)

Customers

Site tal)

ivery r

mer

External B2B Services

- logistics and SCM

- logistics, delivery, …

- supply chain mgmt.

- customer service

urce: Warkentin (2002), p. 274).

M

M

MM

M

MM

M

M

eMerchant Service

B2X Services

* Content Management

* Logistics

* Customer Service

* Shipping

* Marketing

* Financial

* Order Management

B2X Hub

Other B2X Hubs

Enterprises

Transaction XML

Transaction XML

Transaction XML

B2X XML

ERP XML

Proprietary Links

C

C

CC

C

CC

C

C

Internet Business Service

B

B

BB

B

BB

B

B

InternetExchange Exchange

Infrastructure

Fig. 2. Business-to-exchange networks (adapted from: Keenanvision.com).

13280 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

time. Membership in the exchange would provide access to an en-tire range of value-added services, and the site designers couldplug in the desired services via API and URL insertion ‘‘instead ofa custom system integration job’’.

We are witnessing the evolution of agent-based inter-organiza-tional systems that enable complex direct interaction betweenheterogeneous information systems, which allow Web-basedeServices to discover each other, act autonomously, communicateindependently, provide dynamically-configured services to one an-other, and establish composite business systems.

4. Emerging technologies

Concurrent with the rise of the Internet phenomenon, keytechnologies have been developed which enhance the ability ofelectronic commerce oriented firms to analyze data flows and to dis-cover and leverage the resulting information (Chang & Yang, 2009).

4.1. Data warehousing and data mining

A data warehouse is an orderly and accessible repository ofknown facts and related data that are used as a basis for makingbetter management decisions. In a data warehouse, data must beidentified, cataloged, and stored using structures that enable usersto find the correct information when they need it. Data are ex-tracted from operational systems and external information provid-ers, then cleansed, aggregated, and transformed into a databasethat is optimized for decision making (Bischoff & Alexander,1997; Dodge & Gorman, 1998; Hao, Hongwei, & Zili, 2011; Stroulia& Hatch, 2003).

Data mining is the set of activities used to find new, hidden, orunexpected patterns in the warehouse data. A common synonymfor data mining is ‘‘knowledge discovery in databases’’ (KDD).These terms apply to all activities and processes associated withdiscovering useful knowledge from aggregate data. Using acombination of techniques including statistical analysis, neuralnetworks, fuzzy logic, multidimensional analysis, data visualiza-tion, and intelligent agents, KDD can discover both highly useful

and informative patterns within the data that can be used to de-velop predictive models of behavior or consequences in a widevariety of knowledge domains (Gorodetsky, Karsaeyv, & Samoilov,2003; Kaur, Goyal, & Lu, 2011; Kimball, 1996; Letourneau, Famili,& Matwin, 1999; Mavridou, Kehagias, Kalogirou, & Tzovaras,2008; Tanler, 1997).

4.2. Intelligent agents

Intelligent agents play the role of assistants by allowing manag-ers to delegate work to the software agent. Agent technology isfinding its way into many new systems, including decision supportsystems, where it performs many of the necessary decision supporttasks formerly assigned to humans. Software agents are useful inautomating repetitive tasks, finding and filtering information,and intelligently summarizing complex data. Just like their humancounterparts, intelligent agents have the capability to learn fromtheir managers and even make recommendations to them regard-ing a particular course of action (Park & Sugumaran, 2005).

Intelligent agents are ‘‘software entities that have been givensufficient autonomy and intelligence to enable them to carry outspecified tasks with little or no human supervision (Bourdriga &Obaidat, 2004; Nwana et al., 1998).’’ They have also been described(Decker, Pannu, Sycara, & Williamson, 1997; Hendler, 1996; Maes,Guttman, & Moukas, 1999; Park & Sugumaran, 2005; Spector,1997) as programs that act on behalf of their human users to per-form laborious and routine tasks such as locating and accessingnecessary information, resolving inconsistencies in the retrievedinformation, filtering away irrelevant and unwanted information,and integrating information from heterogeneous informationsources (Sugumaran & Davis, 2001; Zaidi, Abidi, Manikam, &Yu-N, 2004). In order to execute tasks on behalf of a business pro-cess, computer application, or an individual, agents are designed tobe goal driven, i.e., they are capable of creating an agenda of goalsto be satisfied. Agents can be thought of as intelligent computer-ized assistants.

Intelligent agent technology can be an important component indata analysis and mining systems (George, 2011). Not only can

M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288 13281

intelligent agents help perform data mining, they can also assist indiscovering, locating, and reporting the most productiveinformation from among vast sources and volumes of data (Chi &Turban, 1995; Klusch, Lodi, & Moro, 2003; Martin, Lakshmi, &Madhusudanan, 2009), relieving managers of this time consumingand burdensome task. The following section describes the architec-ture of a generic agent-based data mining environment and thefunctionality of each of the agents that are part of such a system.

5. Generic architecture for agent-based data mining

While data mining applications are coming to the forefront ofbusiness data analysis and decision making, in order to success-fully execute these applications, a significant amount of a prioriknowledge about the various data mining techniques, their appli-cability to different scenarios and relevant data selection andtransformation is typically required. However, we propose a Gen-eric Agent-Based Data Mining Architecture (GAMA) which hidesmany of the complexities involved in normal data mining opera-tions. GAMA would assist the user in appropriate data selection,data cleansing, data mining method selection and execution, andinterpretation of the results, thereby enabling a casual user inter-ested in applying data mining techniques to decipher trends andbuying behaviors from customer ‘‘digital footprint’’ data. GAMA,shown in Fig. 3, could be tailored to meet the needs of a particularapplication domain. It consists of the following agents: (a) UserInterface Agent, (b) Control Agent, (c) Data-Centric Agent, (d) DataMining Agent, and (e) Visualization Agent. The functions of theagents are described in the following paragraphs.

5.1. User interface agent

The user interface agent provides a graphical interface for theuser to interact with the data mining environment. It containsmethods for inter-agent communication and acquiring user inputand communicating it to the control agent. It also contains scriptsfor dynamically creating HTML documents that convey the resultsof data mining back to the user. The interface agent keeps track ofthe data mining history, user profile, and user preferences.

5.2. Control agent

The control agent is responsible for coordinating the variouslow-level tasks that need to be performed in satisfying a particulargoal. These tasks may include identifying relevant data sources,

User Interface Agent

Corporate Databases/ Data Warehouse

CoA

Database Server 1

Database Server m

Heuristics Ontologies

Corporate Knowledge Sources

Fig. 3. Generic architecture of agent-

requesting services from agents, and generating reports. The se-quence of tasks to be executed is generated from specific ontolo-gies and heuristics using a rule-based approach. This agent alsocontains meta-knowledge about capabilities of other agents inGAMA, available data sources and data warehouses, data miningmethods, etc. The control agent may seek the services of a groupof agents and synthesize the final result.

5.3. Data-centric agent

The data-centric agent actively maintains meta-data informa-tion about each of the existing databases and data warehouses,as well as external data sources. This agent also provides prede-fined and ad hoc retrieval capabilities. It is responsible for retriev-ing the necessary data sets requested by the data mining agent(described below) in preparation for a specific data mining opera-tion. It takes into account the heterogeneity of the databases andresolves conflicts in data definition and representation. The data-centric agent provides facilities for ad hoc and predefined data re-trieval. Based on the user request, this agent generates appropriatequeries and executes the queries against the data warehouse. Theresults are then communicated back to the user or to other agents.

5.4. Data mining agent

The data mining agent is responsible for data cleansing andtransformation, as well as carrying out the data mining operation.This agent may contain specific algorithms that were developed in-house, or it can be the conduit to special purpose data miningproducts that are acquired externally and installed within the envi-ronment. The data mining agent also captures the results of datamining and communicates it to the visualization agent (describedbelow). This agent contains knowledge about the suitability of var-ious data mining methods for different types of problems, inputrequirements for each mining method, format of input data, etc.

5.5. Visualization agent

The visualization agent assimilates the results from the datamining activities and generates the final report that is providedto the user. This agent contains preformatted report templates aswell as a variety of graphical representations such as 3D Bar charts,Association Ball Graph, Rule Graphs, Box Plots, and Trends (Lee &Ong, 1996; Wang & Yi, 2012). Visualization is interactive in thesense that the user can view these graphs and charts from any per-

ntrol gent

Visualization Agent

Data Mining Agent

Data-Centric Agent

based data mining environment.

13282 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

spective by rotating them on screen. This agent stores details aboutreport templates and visualization primitives.

6. Agent-based data mining applications in eCommerce

In this section we specify the role of data mining in the B2C andB2B eCommerce models described earlier. We focus on the flow ofinformation between the various stakeholders and identify oppor-tunities for applying data mining. In particular, we show howGAMA can be customized to support managerial decision-makingand problem solving. As an example, we show how the cross-link-ages between the B2C and B2B models in the adserving industrycan be exploited for the benefit of the companies and individualsinvolved.

6.1. Data mining and intelligent agents in etailing

Web-based B2C retailing (etailing) using posted prices is by farthe most commonly understood and prevalent eCommerce busi-ness model. Pioneers like Amazon.com and eToys.com are at theforefront of the public’s imagination, representing the ‘‘new way’’of going about business in a networked economy. Web-basedshopping technologies give consumers the power to make better-informed decisions that are not constrained by geography, time,or information scarcity. From the seller’s point of view, access tovast global markets and the ability to engage in mass customiza-tion creates myriad new opportunities (Warkentin, Bapna, &Sugumaran, 2000). Amazon’s CEO, Jeff Bezos, wishes to have mil-lions of separate virtual storefronts for its millions of buyers, eachcustomized to the individual’s preferences.

6.2. Agent enhanced etailing architecture

In Fig. 4 we provide a general architecture for etailing that is en-hanced by data mining and intelligent agent technologies. Typi-cally, the front end of a Web-based B2C storefront consists ofmodules for product searching, order management, payment pro-cessing, and customer relationship management. The back endconsists of modules that integrate with enterprise systems suchas inventory control, shipment processing, and order fulfillment.We propose that this architecture be enhanced to include a dataacquisition agent and a data retrieval agent. The data acquisitionagent is responsible for connecting to the enterprise data storageand collecting micro-level ’’clickdata’’ that can subsequently bemined in a variety of ways. The data retrieval agent connects tothe managerial decision support module and supports the knowl-edge discovery process for managers. These enhancements will en-able a consumer Website to automatically identify and supportimportant unique patterns that define the relationship with eachcustomer.

6.3. Data mining in B2B Ecommerce

Web-based exchanges that connect buyers and suppliers in realtime are having a significant impact on procurement and supplychain management. While early EDI technologies connected largebuyers to their suppliers, such networks were proprietary andnon-scalable in nature. Web-based B2B exchanges are designedto include small and medium-sized players as well. The role of datamining and intelligent agents becomes important in this sector asone examines the integration of typical back-end functions likeinventory control with issues of customer service, marketing, andproduction planning. Because of the vast number of buyers andsellers, data-mining agents can play a role in the procurement pro-cess by looking for substitutes, comparing prices, and assessing

supply levels in real time. Inventory controllers can benefit frombetter demand forecasting.

In electronic vertical markets (between businesses within a spe-cific industry), agents have been developed to support the informa-tion exchange function (see Fig. 5). For instance, ApplicationService Providers (ASPs) (such as www.CommerceOne.com), thatbuild and host vertically integrated trading communities on theInternet utilize intelligent agents to aggregate horizontal informa-tion such as shipping patterns. Subsequently, the agent can use thisinformation to negotiate competitive advertising rates withadserver companies such as www.DoubleClick.com and www.ad-force.com. And because a high percentage of corporate procure-ment is regular, routine, repeat business, agents can play aparticularly valuable role to improve the information exchange be-tween the many buyers and sellers in these markets.

We envisage that the current B2B exchanges will become morethan simple market makers that match buyers and sellers. The realbenefits will come when such exchanges support a collaborativeenvironment that includes project management, forecasting, andknowledge sharing within an environment of shared strategic fo-cus. Strategic partners will mutually benefit from the knowledgeacquired through the intelligent mining of their electronically-cap-tured interaction data. Each of these partnership activities addslayers of complexity to the exchange, leading to greater demandfor agent-based data mining technologies.

Finally, future B2B exchanges may employ agent-based datamining to enhance the provision of customized services offeredby the exchange to both buyers and sellers. These include catalogmanagement (conversion, integration, and maintenance), protocoltranslation (for compatibility), sourcing (RFQ, bid coordination),auction management, profile administration, and various supportservices including financing, payment, logistics, tax, and ordertracking.

Beyond B2B exchanges, there are many other models of B2Brelationships, including the adserver network model, in whichadservers act as a broker, bringing together the buyer and sellerof banner ad space, to guarantee that the ad will be displayed orclicked a certain number of times. The adserver sells impressionsto the buyers of ad space and shares this revenue with the ad spaceseller. There are numerous data flows between the individual surf-ing the Web, the adserver, the content-provider Website, and otherWebsites in the adserver’s network. Leading adservers interceptthe user’s click patterns, along with metadata about the content re-quests users make in order to evaluate their surfing pattern. To im-prove the ‘‘stickiness’’ of the content provider site, an adservingcompany like DoubleClick may share knowledge with the providerabout a user’s preferences in order to ensure that the provider’sserver dynamically delivers interesting content to the user. Con-tent manipulation agents can be utilized to dynamically serveappropriate content to each user based on agent-based data min-ing analysis. For example, if a user has recently clicked at anothersite on items related to digital cameras, the adserver may be ableto give the content provider relevant knowledge that leads to serv-ing content related to digital cameras or books about digitalphotography.

Agent-based data mining technology can play a great role in thereal-time analysis of this clickdata and related information. Data isshared by businesses within the B2B component of this relation-ship, and between the user and each of the businesses in this net-work. Managers in these new marketspaces who effectivelyimplement proper agent-based data mining technologies andmechanisms can gain competitive advantage in the race to buildbrand equity and lock in consumer loyalty. The knowledge result-ing from data mining will assist managers in creating intimatecustomer experiences that will ultimately lead to improvedrevenues.

B2B HUB

Supplier 1

Supplier 2

Supplier n Buyer n

Buyer 2

Buyer 1

Collaborative HUB Applications

Supplier Data

Order Data

HUB Data Storage

Buyer Data

Transaction Data

Data Acquisition

Agent

Control Agent

Model Management

Agent

Data Mining Agent

Visualization Agent

User Interface

Agent

Data Warehouse

Data Retrieval

Agent

Managerial Decision Support

. . . .

. . .

.

•Planning & Forecasting •Replenishment •Analytics •Project Management •Knowledge Management

•Auctions •Reverse Auctions •Exchange •Procurement •Catalogs

Fig. 5. Agent-based data mining environement for B2B E-Commerce (Source: Warkentin, Sugumaran, & Bapna, 2001b).

Client WebServer/ E-Store

Customer Data

Order Data

Transaction Data

External Data

Sources

Data Acquisition

Agent Control Agent

Data Warehouse

Model Management

Agent

Log

Enterprise Data Storage

Data Mining Agent

Visualization Agent

Managerial Decision Support

Call Center/Help Desk Service & Support Sales Contact Management Order Checking

Customer Relationship Management (CRM)

Banks E-Cash/Digital Wallet Visa/Mastercard Clearinghouses

Customer Payment Management

Order Management

Order Processing Order Fulfillment Order Tracking

Enterprise Systems

Inventory Distribution Shipment General Ledger

Data Retrieval

Agent

User Interface

Agent

Knowledge Base

Model Base

Fig. 4. Agent enhanced etailing architecture.

M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288 13283

Fig. 6. GAMA user interface screen.

13284 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

7. Implementation of GAMA prototype

Instantiations of a generic agent based data mining tool weredeveloped as demonstration prototypes of the efficacy of this de-sign artifact. The prototype was developed using an ApplicationProgramming Interface (API) connection with eBay, through whichusers can access current and past auction information. While lim-ited to eBay data sources, this system is still able to produceknowledge beyond what is available to users of eBay’s website.An expansion of this tool, if paired with other online data sources

Fig. 7. Category nu

and/or an organizations internal databases, could lead to evenmore in depth knowledge of the domain.

The system is built using eBay’s publicly available SoftwareDevelopment Kit and API interface, in a .NET environment. Connec-tions to the eBay data sources is dependent on a developers ac-count; data requests are limited by type and time period. Fig. 6shows the main screen, where the data source and type of requestcan be selected. Though this demonstration prototype is limited toeBay and Internal Sources, it could easily be expanded as otherdata sources become available.

Menu options allow users to find information on eBay catego-ries (Fig. 7), list the auctions in a category with various filtersand sorting options available (Fig. 8), or search using keywordsand other criteria for current auctions (Fig. 9). Furthermore, abuyer evaluator allows the user to see other auctions that a userhas been involved in, as well as how much money he or she has re-cently spent (see Fig. 10).

The category number can be used in a number of informationrequests to limit the results to only certain segments of eBay auc-tions. There are general categories for large classes of products andseveral layers of more specific categories under them. The categoryused throughout the examples below is 177, for PC based laptopcomputers, which is under the 58058 category of computers andnetworking category.

Once the user has found a category number of interest, he or shecan list all the auctions currently available in that category. Fig. 8shows the listing for category 177, along with additional filtersfor the type of auctions and sort options. Results can also be lim-ited to the region where the auction is originating. This data couldbe used to set up weekly or daily reports, detailing the auctionscoming up that match the defined criteria the user is interestedin. An additional agent implementation could be designed to auto-matically bid on auctions matching a pre-defined set of rules.

Another way to find specific auctions is the keyword searchinterface that allows the user to submit search terms and view

mber listing.

M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288 13285

the auctions matching those words. There are also options for fil-tering results based on price range, on whether PayPal is accepted,or on the category the auction is listed under. There are also a vari-ety of options for the sorting of results. Again, these searches couldbe developed to run periodically and highlight certain auctions forfurther examination, or an automated bid could be submitted forsome auctions.

The final example of an implemented data agent is the buyerevaluator, which examines a user’s feedback and helps determinehow viable the user is as a buyer. As eBay does not require any-thing more than an active email address to establish an account,there can be many accounts created and sometimes abuses occur.For some high value items, there can be a number of unqualifiedbidders who have no intention of buying a product. There is alsothe danger of fraud, where a buyer convinces a seller to ship anitem before it has been paid for. The eBay web page will let userssee the feedback for other users, but it is difficult to see how muchanother user is really buying. The agent below is built to look upnot just the feedback received by a user, but also the amount ofmoney spent by the user on those auctions. A larger amount paidand positive feedback indicates that the user is more serious aboutan auction and more likely to pay for the item if won.

While limited, this instantiation of the GAMA architectureshows that implementation of the concepts is possible and thatvaluable knowledge can be obtained even with a limited set ofdata. Adding more sources and more functionality to this systemwould increase its usefulness and value. As more data becomesavailable through networked connections, the need to automate

Fig. 8. Listing of aucti

more parts of the process will grow. Also, the addition of internalsources to those shown would allow for the combination of dataand an expansion of the possible data analysis.

8. Managerial implications and future directions

Many companies in both B2C and B2B markets are implement-ing agent-based data mining technologies. Firms using data miningprocesses within these eCommerce marketspaces should considersome key managerial implications. Success will be determined bythe way these tools are used, not by the tools themselves.

First, many firms have not yet effectively implemented datamining technology and are failing to collect valuable data everyday. This valuable data, which leads to an understanding of a com-pany’s market, may make the difference between long-term suc-cess and failure in these competitive marketspaces. Data miningtechniques provide important tools for personalizing a customer’sshopping experience and creating customer intimacy in an onlineexperience. Managers of eCommerce sites must develop and useagent-based data mining to maximize their chance of success inthese marketspaces.

Managers must also develop realistic business rules for evaluat-ing and interpreting data mining results. These business rules willbe unique to each company’s value proposition and customer base.After interpreting the results, managers must implement suchinformation effectively. Many data mining efforts have incorpo-rated state-of-the-art methods but have not led to success due to

ons in a category.

Fig. 9. Keyword search of listings.

Fig. 10. Buyer evaluation.

13286 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288 13287

poor implementation protocols. It is imperative that managersorganize the results of the knowledge discovery process so thatthe exercise leads to appropriate managerial responses.

Consumer privacy is another important managerial issue forcompanies pursuing agent-based data mining. The large amountof clickdata which identifies patterns of usage by individual cus-tomers can be combined with personally identifiable information(such as name, address, and telephone number) to generate pro-files, creating significant opportunity for abuse (Harmon, DeLoach,& Robby, 2009; Junnarkar, 2000).

In order to ensure that agents maximize their ability to gathervaluable data for mining into managerial knowledge, it is impera-tive that the players in these electronic marketspaces adopt XML-based standards for representing attributes and attribute values oftheir products, services, sales, customers, and policies (Warkentin,Sugumaran, & Bapna, 2001a). Interorganizational systems cannothave a significant impact unless a standard data representationscheme is used for data of mutual value. The true potential of intel-ligent agents to efficiently exchange information will not be un-locked unless and until there is a common standard for therepresentation of all product and service attributes which can beeasily transferred and interpreted by all economic players acrossthe Internet. An international standardized data representationscheme for product and service attributes would extend the capa-bilities of agent-based data mining processes, thus further improv-ing the efficiency of all marketspaces throughout the World WideWeb.

The emergence of data mining and intelligent agents at thesame time that millions of individuals have gone online to pur-chase from thousands of new Websites has created an excitingopportunity for practical technological convergence. As eCom-merce moves many business processes and activities online, newdata streams are born and a chance for greater efficiencies is gen-erated for those willing to carefully perform the correct data anal-ysis procedures. Agent-based data mining will enable firms tocapture the full potential of this technological convergence.

Acknowledgments

Dr. Sugumaran’s research has been supported by Sogang Busi-ness School’s World Class University Program (R31-20002) fundedby Korea Research Foundation and the Sogang University ResearchGrant of 2011.

References

Bischoff, J., & Alexander, T. (1997). Data warehouse: Practical advice from the experts.New Jersey: Prentice Hall.

Bourdriga, N., & Obaidat, M. S. (2004). Intelligent agents on the web: A review.Computing in Science and Engineering, 6(4), 35–42.

Carr, N. G. (2000). The future of commerce – Hypermediation: Commerce asClickstream. Harvard Business Review, Jan/Feb.

Chaimontree, S., Atkinson, K., & Coenen, F. (2012). A multi-agent based approach toClustering: Harnessing the power of agents. In L. Cao et al. (Eds.). Lecture Notesin Computer Science (Vol. 7103, pp. 16–29). Berlin Heidelberg: Springer Verlag.

Chan, P. K., Fan, W., Prodromidis, A. L., & Stolfo, S. J. (1999). Distributed data miningin credit card fraud detection. IEEE Intelligent Systems and their Applications,14(6), 67–74.

Chang, Y. C. & Yang, P. C. (2009). Ontology-based intelligent web mining agent forTaiwan travel. In IEEE/WIC/ACM international joint conference on web intelligenceand intelligent agent technology (IAT’09) (pp. 421–424).

Chi, R. T., & Turban, E. (1995). Distributed intelligent executive information system.Decision Support Systems, 14, 117–130.

Cunningham, M. J. (2001). Maximizing the power of partners. eBusiness Advisor,19(1), 32–41.

Decker, K., Pannu, A., Sycara, K., & Williamson, M. (1997). Designing behaviors forinformation agents. In Proceeding of Autonomous Agents’97 Feb 5–8, Marina delRey, CA (pp. 404–412).

Dodge, G., & Gorman, T. (1998). Oracle8 data warehousing. New York: John Wiley &Sons.

Gannon, T., & Bragger, D. (1998). Data warehousing with intelligent agents.Intelligent Enterprise, 1(1), 28–37.

Gao, J., Yan, T., & Dang, Y. (2012). A web log mining system based on MDP algorithm.In L. Cao et al. (Eds.), Recent advances in computer science and informationengineering (pp. 447–453). Berlin Heidelberg: Springer Verlag.

George, A. (2011). Intelligent agent based architecture for patient monitoring in biosensor networks. In V. V. Das & N. Thankachan (Eds.). Communications incomputer and information science (Vol. 250, pp. 180–186). Berlin Heidelberg:Springer-Verlag.

Gomes, M. F., & Canuto, A. M. (2006). Carcara: A multi-agent system for web miningusing adjustable user profile and dynamic grouping. In IEEE/WIC/ACMinternational conference on intelligent agent technology (IAT’06) (pp. 187–190).

Gorodetsky, V., Karsaeyv, O. L., & Samoilov, V. (2003), Multi-agent technologyfor distributed data mining and classification. In IEEE/WIC InternationalConference on Intelligent Agent Technology (IAT’03), October 13–17, Halifax,Canada (p. 438).

Gorodetsky, V., Karsaev, O. L., Samoylov, V., & Serebryakov, S. (2008). Interaction ofagents and data mining in ubiquitous environment. In IEEE/WIC/ACMInternational Conference on Intelligent Agent Technology (IAT’08) (pp. 562–566).

Grimes, S. (1998). Agents come in from the cold. Database Programming and Design,11(4), 48–53.

Hao, Y., Hongwei, S., & Zili, Z. (2011). The application of e-commerce system basedon data warehouse. In IEEE joint international information technology andartificial intelligence conference, Aug 20–22, Chongqing, China (pp. 493–496).

Harmon, S., DeLoach, S., & Robby (2009). Abstract requirement analysis inmultiagent system design. In IEEE/WIC/ACM International Joint Conference onWeb Intelligence and Intelligent Agent Technology (IAT’09) (pp. 86–91).

Hendler, J. (1996). Intelligent agents: Where AI meets information technology. IEEEExpert, 11(6), 20–23.

Holmes, D. E., Tweedale, J. W., & Jain, L. C. (2012). Advances in Intelligent DataMining. In J. Kacprzyk & L. C. Jain (Eds.), Data Mining: Foundations and IntelligentParadigms (pp. 1–7). Berlin Heidelberg: Springer Verlag.

Hong, H. (2010). Study on the integration of supply chain management in retailbusiness. In International conference on e-product e-service and e-entertainment(ICEEE), Nov. 7–9, Henan, China (pp. 1–4).

Jain, V. (2012). Information retrieval through multi-agent system with data miningin cloud computing. International Journal Computer Technology & Applications,3(1), 62–66.

Jayabrabu, R., Saravanan, V., & Vivekanandan, K. (2012). A framework: Clusterdetection and multidimensional visualization of automated data mining usingintelligent agents. International Journal of Artificial Intelligence & Applications(IJAIA), 3(1), 125–138.

Jiang, Y., & Zhao, J. (2012). An empirical research of the creative process ofcollaborative e-business capability in service industry. International Journal ofNetworking and Virtual Organizations, 10(1), 73–87.

Junnarkar, S. (2000). DoubleClick accused of Unlawful Consumer Data Use. <http://www.news.cnet.com/2100-1023-236216.html> [accessed 14.03.2012].

Kaur, P., Goyal, M., & Lu, J. (2011). Data mining driven agents for predicting onlineauction’s end price. In IEEE symposium on computational intelligence and datamining (CIDM), April 11–15, Paris, France (pp.141–147).

Kehagias, D. D., & Mitkas, P. A. (2007). Efficient e-commerce agent design based onclustering ebay data. Web Intelligence/IAT Workshops, 2007, 495–498.

Khalil, W., & Shikuta, E. (2012). Virtual organization for computational intelligence.Studies in Computational Intelligence (Vol. 396). Berlin Heidelberg: Springer, pp.437–464.

Kimball, R. (1996). The data warehouse toolkit: Practical techniques for buildingdimension data warehouses. New York: John Wiley.

Klusch, M., Lodi, S., & Moro, G. (2003). The role of agents in distributed data mining:issues and benefits. In Proceedings of the IEEE/WIC international conference onintelligent agent technology (p. 211).

Lee, R. S. T., & Liu, J. N. K. (2004). IJADE web-miner: An intelligent agent frameworkfor internet shopping. IEEE Transaction on Knowledge and Data Engineering,16(4), 461–473.

Lee, H., & Ong, H. (1996). Visualization support for data mining. IEEE IntelligentSystems & their Applications, 11(5), 69–75.

Letourneau, S., Famili, F., & Matwin, S. (1999). Data mining to predict aircraftcomponent replacement. IEEE Intelligent Systems and their Applications, 14(6),59–66.

Li, C., & Li, K. (2011). A practical framework for agent-based hybrid intelligentsystems. In Seventh international conference on computational intelligence andsecurity (CIS), Dec. 3–4, Hainan, China (pp. 199–203).

Liao, H., Yang, L., & Geng, X. (2011). Application research of new distributed datamining model based on intelligent agent and web service in digital gas fields. InInternational conference on computational and information sciences (ICCIS), Oct.21–23, Chengdu, China (pp.137–140).

Liu, B., Cao, S. G., & He, W. (2011). Distributed data mining for e-business.Information Technology and Management, 12(2), 67–79.

Maes, P., Guttman, R. H., & Moukas, A. G. (1999). Agents that buy and sell.Communications of ACM, 42(3), 81–87. 90–91.

Marik, V., & McFarlane, D. (2005). Industrial adoption of agent-based technologies.IEEE Intelligent Systems, 20(1), 27–35.

Martin, A., Lakshmi, T. M., & Madhusudanan, J. (2009). Multi agent communicationsystem for online auction with decision support system by JADE and TRACE.Journal of Convergence Information Technology, 4(2), 154–163.

Mavridou, E., Kehagias, D. D., Kalogirou, K., & Tzovaras, D. (2008). An agent-orienteddata mining framework for mass customization in the automotive industry. InIEEE/WIC/ACM International Conference on Intelligent Agent Technology (IAT’08)(pp. 575–578).

13288 M. Warkentin et al. / Expert Systems with Applications 39 (2012) 13277–13288

Moemeng, C., Zhu, X., Cao, L., & Jiahang, C. (2010). i-Analyst: An agent-baseddistributed data mining platform. In IEEE international conference on data miningworkshops (ICDMW), Dec. 13, Sydney, Australia (pp.1404–1406).

Mowshowitz, A. (1997). A virtual organization. Communications of the ACM, 40(9),30–37.

Nassiri, N. (2009). Software agent in desktop virtual shopping. In IEEE/WIC/ACMInternational Joint Conference on Web Intelligence and Intelligent Agent Technology(IAT’09) (pp. 618–620).

Nwana, H., Rosenschein, J., Sandholm, T., Sierra, C., Maes, P., & Guttmann, R. (1998).Agent-mediated electronic commerce: issues challenges and some viewpoints.In Proceedings of the 2nd International Conference on Autonomous Agents, May 9–13, Minneapolis, MN (pp. 189–196).

Park, S., & Sugumaran, V. (2005). Designing multi-agent systems: A framework andapplication. Expert Systems with Applications, 28(2), 259–271.

Rajan, J., & Saravanan, V. (2008). A framework of an automated data mining systemusing autonomous intelligent agents. In International conference on computerscience and information technology (pp. 700–704).

Rao, V. S. (2010). Multi agent-based distributed data mining: An overview.International Journal of Reviews in Computing, 3, 83–92.

Spector, L. (1997). Automatic generation of intelligent agent programs. IEEE Expert,12(1), 3–4.

Stroulia, E., & Hatch, M. P. (2003). An intelligent-agent architecture for flexibleservice integration on the web. IEEE Transactions on Systems, Man andCybernetics: Part C, Applications and Reviews, 33(4), 468–479.

Sugumaran, V., & Davis, K. (2001). A natural language-based multi-agent system forlegal research. In Proceedings of the 2001 Americas conference on informationsystems (AMCIS) (pp. 412–414).

Sugumaran, V., & Bose, R. (1999). Data analysis and mining environment: Adistributed intelligent agent technology application. Industrial Management andData Systems, 99(2), 71–80.

Tanler, R. (1997). The intranet data warehouse. New York: John Wiley & Sons Inc..Tuzhilin, A. (2012). Customer relationship management and web mining: The next

frontier. Data Mining and Knowledge Discovery, 24(3), 584–612.

Wang, X., & Yi, B. (2012). The application of data cubes in business datavisualization. Computing in Science & Engineering, 14, 1–10.

Warkentin, M., Bapna, R., & Sugumaran, V. (2000). The role of mass customization inenhancing supply chain relationships in B2C E-commerce markets. Journal ofElectronic Commerce Research, 1(2), 1–17.

Warkentin, M., Sugumaran, V., & Bapna, R. (2001a). Intelligent agents for electroniccommerce: Trends and future impact on business models and markets. In R. S.Mahbubur & R. J. Bignall (Eds.), Commerce and software agents: Cases technologiesand opportunities (pp. 101–120). Hershey, PA: Idea Group Publishing.

Warkentin, M., Sugumaran, V., & Bapna, R. (2001b). EKnowledge networks for inter-organizational collaborative ebusiness. Logistics Information Management, 14(1/2), 149–162.

Warkentin, M. (2002). Dynamic digital process integration in business-to-businessnetworks. In M. Warkentin (Ed.), Business to business electronic commerce:Challenges and solutions (pp. 261–281). Hershey, PA: Idea Group Publishing.

Willow, C. C. (2005). Automatic agent and data mining for e-commerce. InInternational conference on information technology: Coding and, computing(ITCC’05) (pp. 380–384).

Xu, F. J., Tong, F. C. H., & Tan, C. J. (2011). Auto-ID enabled tracking and tracing datasharing over dynamic B2B and B2G relationships. In IEEE internationalconference on RFID-technologies and applications, Sitges, Sept 15–16 (pp. 394–401).

Yarom, I., Rosenschein, J. S., & Goldman, C. V. (2003). The role of middle-agents inelectronic commerce. IEEE Intelligent Systems, 18(6), 15–21.

Zaidi, S. Z. H., Abidi, S. S. R., Manikam, S., & Yu-N, C. (2004). ADMI: A multi-agentarchitecture to autonomously generate data mining services. In Second IEEEinternational conference on intelligent systems (pp. 273–279).

Zhang, H., Wang, L., & Shen, G. (2012). Data exchange standard on the supply chainbased on XML. Advances in Intelligent and Soft Computing (Vol. 136). BerlinHeidelberg: Springer, pp. 379–386.

Zuo, W., & Hua, Q. (2012). The application of web data mining in the electroniccommerce. In Fifth international conference on intelligent computation technologyand automation (ICICTA), Jan 12–14, Zhangjiajie, Hunan (pp. 337–339).