INTEGRATING MOBILE AGENT AND CONTEXT-AWARE

WORKFLOW ANALYSIS FOR M-COMMERCE APPLICATIONS

Jiazao Lin

, Xining Li

and Lian Li

Department of Computer Science & Department of Mathematics, Lanzhou University, Lanzhou, Gansu, P. R. China

School of Computer Science, University of Guelph, Guelph, Ontario, Canada

Keywords: M-commerce, Mobile Agent, Context-aware Workflow, Service Discover.

Abstract: Mobile commerce (M-commerce) is an attractive research area due to its relative novelty, rapid growth, and

great potential in business applications. Unfortunately, there are a number of constraints effecting both

performance and usability of mobile devices and network bandwidth. In addition, existing M-commerce

applications are lack of fully automated business processes and still require significant manual effort. In this

paper we present a general solution of integrating mobile agent and context-aware workflow to implement

automated trading task and compose services dynamically in real time to create a highly personalized

assistant. Furthermore, the proposed context-aware model derives from a set of ontology of descriptive

contextual attribute for knowledge sharing and logical inference. Instead of executing a fully defined

process description, the composite workflow can be executed with forward or backward selection of

services determined at run time. We have carried out an evaluation experiment. The results show that our

proposed solution is feasible and viable.

1 INTRODUCTION

With the development of Internet and its related

technologies, the most significant change of our

daily lives is the way of conducting business. There

is no doubt that Electronic commerce (E-commerce)

is the most successful model to explore opportunities

and expand business into global commercial market.

It offers new channels and business models for

buyers and sellers to effectively and efficiently trade

goods and services over the Internet （Laudon and

Traver, 2006）. However; the traditional E-commerce

is based on the client/server approach, which

requires a stable connection between client and

server. Such a requirement makes some restrictions

on the spatial and temporal activities. With various

types of Internet-enabled mobile devices, such as

PDA’s, mobile phones, pocket PC’s, etc., the mobile

Internet is opening the door to numerous new mobile

applications and services that will assist mobile users

to engage in time-critical, goal-driven tasks (Sadeh,

2002). M-commerce has emerged and attracted a

growing number of research efforts recently. The

basic idea of M-commerce is to conduct business

transaction with mobile devices and

telecommunication/wireless networks, either directly

or indirectly (Bai and et al., 2005). It can help

nomadic users to roam a wide range of services and

products over the Internet on an anywhere and

anytime basis. Due to the mobility, personality and

flexibility, M-commerce is likely to become the

main business model in the near future (Ngai and

Gunasekaran, 2007).

One issue which must be considered in the

design of M-commerce applications is the limitation

of mobile environment and interface between users

and applications. In the mobile environment, it is

impossible to retain a long connection between the

client and service providers (Qiang and Hin, 2002)

and it is also unreliable to transfer a huge amount of

data between client and server. Comparing to

applications based on desktop, the mobile handhold

devices have some physical constraints, such as

small screen size, limited battery capability, limited

storage and computing capacity, and low-bandwidth

links with a speed which is slow and varies for

different periods of time. In addition, the nomadic

users need to frequently check trading opportunity

(Mihailescu and et al., 2002), as well as carry out

fuzzy and complex information exchange and

decision-making tasks. Therefore, it leads to the

raise of revenue and creates the risk of missing trade

opportunities if the trade time is constrained by the

109

Lin J., Li X. and Li L. (2010).

INTEGRATING MOBILE AGENT AND CONTEXT-AWARE WORKFLOW ANALYSIS FOR M-COMMERCE APPLICATIONS.

In Proceedings of the International Conference on e-Business, pages 109-115

 SciTePress

limited availability of physical access to the service

(Kowalczyk and et al., 2003).

Furthermore, It is well-known that an

M-commerce transaction involves a sequence of

activities, such as negotiation, purchasing, shipment,

payment and logistic services. These activities

demand additional features of automation and

optimization. Existing M-commerce applications are

lack of fully automated business processes and still

require significant manual effort.

In order to solve the problems mentioned above,

we propose the concept of integrating mobile agent

and context-aware workflow to implement

automated trading task, and compose services

dynamically in real time to create a highly

personalized assistant. A mobile agent is a

self-contained executable entity which is capable of

autonomously roaming the Internet to access

computing and information resources to carry out

user specified tasks. Deploying mobile agents in

M-commerce could add automatic and intelligent

capabilities to conduct a business transaction (Bădică

and et al., 2005), and offers mobile users the

freedom of connection or

disconnection/reconnection functionality to reduce

network cost and power consumption. Workflow is

the process within a system and the rate at which that

happens. Through workflow analysis, we could

observe and extract how this process takes place and

improve it for efficiency and effectiveness. To

achieve user desired goals, a management

mechanism can be used to aggregate different

business steps into workflows, and automatically

chain multiple services together by using planning

(Chakraborty and et al., 2005). However, existing

workflow management systems lack an appreciation

of the content of a business process and do not make

decisions based on the nature of information being

gathered, that is, many decisions are traditionally

made in the process description at design time. In

our system, we propose to extend workflow systems

to integrate and utilize contextual information

relevant to nomadic user to enhance a higher level of

automation. For an M-commerce application, the

contextual information refers to nomadic user profile

and preference which plays a crucial role in the

simplification of the interaction between human and

the virtual digital world. A general sense of context

awareness refers to the ability of an application to

discover and take advantage of contextual

information (Dey and Abowd, 1999). In our work

the context information involves the Person Profile,

Environment Profile, Current Activity and Context

History. The context-awareness and adaptability

accommodated into the workflow is called

Context-aware workflow, which is defined as the

process of autonomous and adaptive constructing

from atomic services to form a specific, complex

task with contextual information. The objective of

our research is to design a flexible, adaptable mobile

infrastructure to accommodate M-commerce

applications.

The reminder of this paper is organized as

follows. In Section 2, we will discuss some highlight

concepts and methodologies involved in our system,

such as mobile agent, context model and

context-aware workflow. Section 3 will present our

proposed solution and the layered system

architecture. In Section 4, we show an experiment to

testify the feasibility of the proposed architecture.

Finally, we conclude our work and indicate future

works.

2 HIGHLIGHT CONCEPTS

AND METHODOLOGIES

The dramatic evolution of wireless/telecommunication

technologies and mobile computing devices has

attracted many researchers to migrate their interests

from E-commerce to M-commerce. Several

M-commerce platforms have been proposed which

use either traditional client-server model or

agent-based architecture, such as eAuctionHouse

(Sandholm and Huai, 2000) and MAE (Mihailescu

and et al., 2002). Other M-commerce researches

involve mobile advertising, mobile stock trading,

mobile marketing, mobile content distribution, etc.

On the other hand, in order to explore automation

and optimization of business processes, a number of

context-aware applications have been developed to

support service composition, such as CAWE

(Ardissono and et al., 2007) and CACS (Luo and et

al., 2006). However, due to the varity of mobile

devices, evolving wireless and telecommunication

technologies, heterogeneous platforms and existing

and emerging business models, there is still a long

way to go in terms of developing the user friendly

M-commerce applications. Therefore, in our

proposed infrastructure, it will incorporate the

concepts of mobile agent, context aware computing,

web services, workflow, as well as their related

models and methodologies. We will present a brief

overview of these researches, and discuss their

relevance with our system.

2.1 Mobile Agent Technology

Agent technology is a paradigm for structuring,

designing and building systems that require complex

interactions between autonomous distributed

ICE-B 2010 - International Conference on e-Business

110

components. The agent model has been recognized

as a highly effective implementation technique in

E-commerce or M-commerce (Kotz and Gray,

1999). Especially the mobile agent paradigm has

been deployed as a good candidate to overcome the

limitations of connectivity, latency and bandwidth of

wireless and telecom networks (Hagen and et al.,

1998). For example, a nomadic user can dispatch

mobile agents from a handhold device to perform an

M-commerce application. Once the application has

been launched, the user may disconnect from the

network. The execution results can either be sent

back by mobile agents through SMS/email, or be

collected when the user receives a notice and

reconnects to the network. In order to ease the access

and participation, reduce costs and improve trading

efficiencies, support for automation of

decision-making, our proposed system adopts the

agent technology and deploys two kinds of agents,

namely, stationary agent and mobile agent. A

stationary agent always resides at its host and be

classified as a home agent or a vendor agent. A

home agent is responsible to accept requests from a

mobile user and dispatch corresponding mobile

agents to invoke a trade transaction, whereas a

vendor agent acts as the representative of the vendor

to keep track of all transactions, inquires, and

possible trade negotiation. A mobile agent represents

the “runner”, who roams the Internet to carry out the

assigned task. Obviously, mobile agents must be

lightweight in order to swiftly move across the

network. Mobile agents are initialized and

dispatched by its home agent to migrate to multiple

provider sites and communicate with vendor agents

to perform their tasks. Having finished their assigned

work, mobile agents may move back to the home

server to deliver the execution results.

2.2 Context Model

M-commerce applications have a great demand for

context awareness, that is, a need to exploit various

information in order to adapt application behaviors.

Most existing applications focus on location

awareness, i.e., to provide personalized services

based on the customer’s current position in physical

space. Generally speaking, context could be any

information that is helpful to characterize the

situation of an entity, where an entity can be a

person, a place, a physical or a computational object

(Dey, 2001). In addition, context could be either

explicitly indicated by the user or implicitly

extracted from other information sources. Certainly,

mobile customers want to find the best deal in an

M-commerce environment. The best deal can only

be obtained by appropriately combining information

gathered from various shopping services in the

physical vicinity. To facilitate the development of

extensible and interoperable context aware

applications and make contextual data usable and

sharable by M-commerce applications, it is essential

to have a set of principles for specifying any given

context from any domain. To achieve this, a set of

well-defined, uniform context models and protocols

is required. As a formal representation of entities,

ideas, and events, along with their properties and

relations within a system of categories, ontology

allows sharing a common understanding of

information and deriving additional information

from what is already known (Uschold and

Gruninger, 1996).

Derived from above discussion, we consider the

context as the entire collection of entities and their

properties that can form a meaningful relationship

between mobile users and M-commerce applications.

As a result, we define the context dimensions which

have to be considered in our system. It consists of a

set of elements along four axes, namely, Person

profile (Name, Sex, Nationality, Birthday, Social

Role, Address and Phone, Friend List), Environment

profile(Current Time, Date, Longitude, Latitude,

Weather, Temperature, Light, Noise), Context

history (Visited Webs and Shopping History) and

Current activity. It is worth to note that we only

listed a portion of context in our system, the content

will extended along with the progress of system

development.

2.3 Workflow Management

In M-commerce applications, it is desirable that

business activities to be completed quickly with high

quality and low cost. A well-designed workflow

management system can provide potential

competitive advantage to manipulate a series of tasks

within a business transaction to produce a final

result. In recent years, web services are growing and

evolving rapidly in M-commerce applications. The

development of new service composition by

integrating existing services is generating

considerable interests in business communities.

There are several typical approaches to the web

services composition (Milanovic and Malek, 2004).

However, these conventional workflow management

models and systems do not provide sufficient

characteristics such as automatic composition and

adaptability verification. It is therefore important to

study how to design a workflow management system

which can integrate and utilize context information

in the analysis process and activity enactment.

For this purpose, we propose a context-aware

workflow management system, as an autonomous

INTEGRATING MOBILE AGENT AND CONTEXT-AWARE WORKFLOW ANALYSIS FOR M-COMMERCE

APPLICATIONS

111

and adaptive process which is able to construct a

complex service from atomic activities to achieve a

specific task with contextual information. The

composition service should be executed with the

freedom of forward or backward selection and the

service selection is determined at run time instead of

executing a fully defined process description.

Therefore, we introduce an abstract hierarchy whose

higher-level elements describe the tasks to be

performed in a generic way. The hierarchy consists

of two layers, namely, Abstract Activity and Activity

Flow Description (AFD).

An Abstract Activity is a high-level description of

the capabilities and categorization of an atomic

service with similar functionality. The service

functionality is specified in terms of its inputs,

outputs, preconditions and effects. Each abstract

activity has an associated set of context-dependent

implementations representing the alternative courses

of the action which the service management model

should actuate, depending on the context states. Each

Abstract Activity is associated with a service type.

There are two types of concrete services in our

system, namely, E-service and A-service. An

E-service represents a common web service which

includes the service provider, the purpose of the

service and the method of invocation. On the other

hand, an A-service indicates an agent-based service

which includes the remote agent host, name of the

vendor agent, and the method of agent

communication. To invoke an A-service, a mobile

agent must migrate to the remote agent server and

communicate with the named vendor agent. A

vendor agent is a persistent agent acting as the

representative of the vendor. It provides services

interactively with the mobile agent to implement

authentication, query, negotiation, etc.

An AFD indicates the order of a collection of

activities without the details of execution. It may

contain one or more abstract activities. An abstract

activity is replaced by one of the concrete

implementations at run time. An AFD can also be

predefined, to customize the services requested by

the nomadic users. It is the responsibility of

workflow management system to determine the

actual implementation at execution time based on the

context information. Therefore, it is necessary to

deploy a conceptual level specification language to

describe process product, service and information

flows, including the tasks, the dependencies between

the tasks and the required roles. In our system, we

will employ XML to describe a logical business

transaction. The primary reason is the flexibility

offered by XML in terms of structured multi-object

documents, compact message construction as well as

the wide acceptance of XML as the communication

standard for wireless-based applications.

2.4 Service Discovery

Service discovery has been widely studied to allow

automatic detection of services, especially web

services offered by various service providers. A web

service can be invoked by using a specified protocol,

such as Simple Object Access Protocol (SOAP), and

has an interface described by Web Service

Description Language (WSDL) and its related

information is published to the Universal

Description, Discovery and Integration (UDDI). A

service discovery protocol mainly involves

dynamically discover and select the best currently

available service that fits the need of a specific

requirement from user. At this stage, we do not

consider the selection rules and selection policies,

instead, we focus our attention on a novel approach

to discover and select service on the basis of

Abstract Activity and contextual information. To get

the benefits of the web service standardization and to

avoid the redesign of another lookup service we

adopt UDDI registry with some extension.

Concretely speaking, a new tag <AgentService> is

added to a UDDI service registry if the service is an

A-Service, that is, the service is implemented by

agent-agent interaction.

In this stage, we will illustrate a typical control flow

of service discovery and service selection during

execution of an M-commerce transaction. To begin

with, a lightweight mobile agent is dispatched by the

mobile device to deliver a XML-based AFD

message to the Home server. Having received the

AFD message, the Home Agent will transfer the

message to the workflow management system for

further processing. The Service Discovery and

Selection module extracts atomic activities from the

AFD message and uses system defined selection

rules and contextual information to search available

services from the UDDI server. Based upon the

search results, the workflow management system

will hook up a concrete implementation to each

atomic activity, that is, to invoke a web service if the

result returned by discovery module is of type

E-service, or to create a mobile agent if the result is

of type A-service. An E-service will be invoked by

the standard SOAP protocol, while an A-service is

carried out by a mobile agent who migrates to one or

more remote hosts and communicates with various

vendor agents to obtain required services.

ICE-B 2010 - International Conference on e-Business

112

3 OVERVIEW OF PROPOSED

SYSTEM ARCHITECTURE

To support our proposed approach, we are now in

the stage of developing the workflow management

prototype system. The main characteristics of the

prototype design are adaptability and flexibility.

Adaptability is achieved by the context awareness

mechanism embedded in the system. Flexibility is

obtained by the layered architecture, as presented in

Figure 1, where each layer will be wrapped by web

service interface. The top layer consists of two

function modules which reside in the mobile device.

Portal: This module constitutes the interface of the

mobile user and the handhold device. It mainly

encompasses different GUI facilities where the user

can customize tasks by setting preferences,

permission profile and personal information. There

are three major functions in this module: Service

Information model aggregates a series of abstract

activities with similar functions in M-commerce

applications. Activity Design GUI is a tool for the

mobile user to customize an M-commerce

application by setting requirements, preferences, and

permission profile. User Information Panel offers

functionality to manage user’s personal information

and contextual information.

Abstract and Connection: This module serves as the

communication and operation bridge between the

mobile user and the home server and includes two

major functions. Abstract Composition Engine is

responsible to extract abstract activities from user’s

specification and transforms the abstract activities to

an AFD message. Transform Model Engine is used

to transform concrete services into abstractions

which are cached in Abstract Activity Centre.

The middle layer is the core of our system, that is,

the Context-aware Workflow Management, which

involves three main modules:

Service Composition Engine: This module provides

the function of translating the well-defined abstract

Description Level flow into a concrete workflow in

which the required resources will be bound. It

manages, controls the workflow, and designates the

appropriate service to accomplish the task.

Workflow Translation function interprets and

decomposes the AFD message came from the top

layer into a workflow. Logic Control Repository

specifies the business logic of the workflow. XML

parser is used to parse the AFD message. Service

Management maintains service patterns and

templates which are frequently used in M-commerce

applications. Service Discovery and Selection

communicates with UDDI servers, evaluates

application conditions for each candidate service

Figure 1: Overview of the System Architecture.

instance and matches appropriate service to an

atomic activity.

It will consult Context Manager to bind the

context-dependent information to each concrete

service implementation.

Agent Management Engine: This module is

supported by the IMAGO system [X. Li, 2006].

Context Management Engine: The major function of

this module is to gather and process contextual

information. History Profile Repository handles

every activity performed by the nomadic user across

a time span. Context Repository stores all

information related to the mobile user and the user’s

environment. Context Aggregator is used to create

new context space based on the existing knowledge

and updates. Profiles and Preference Management

provides the function of managing the explicit user

profile and interest information in a canonical

method. Profiles and Preference Learning

automatically checks and updates context

information through a learning algorithm, such as

services and web sites visited frequently.

The bottom layer constitutes the Physical Execution

Environment of the home server. In fact, it is a

virtual machine specially designed for M-commerce

applications.

4 A PROTOTYPE EXPERIMENT

The design methodology and system architecture

discussed in previous sections are our ongoing

INTEGRATING MOBILE AGENT AND CONTEXT-AWARE WORKFLOW ANALYSIS FOR M-COMMERCE

APPLICATIONS

113

research project. In order to verify the feasibility of

our design, we have conducted some preliminary

experiments of system components and functions

that are considered to be critical to the viability of

our approach. We will use a simple example to

demonstrate our experiment. Suppose that a mobile

user wants to arrange a trip from Beijing to

Shanghai. The trip schedule contains booking a

one-way flight ticket, arranging airport pickup,

making a hotel and a restaurant reservation. The trip

reservation starts at the handhold device of the

mobile user. At this stage, we have a simple GUI

portal. It should be noted that the prototype at

current phase is neither complete nor user friendly. It

only serves as a meaningful means with visual

representation and guidance to the mobile users for

initiating an M-commerce application. In order to

cope with context information and context

reasoning, we adopt Jena Semantic Web Toolkit to

simulate the workflow management module. In the

prototype experiment, we defined four classes to

simulate and verify the service discovery and

activity execution.

Through this experiment, we believe that our

proposed approach is capable of adapting existing

techniques, such as web service, service discovery,

etc., and generating sufficient information for

context-aware workflow analysis.

5 CONCLUSIONS

In this paper, we presented the design and

architecture which integrates mobile agent

technology and context aware workflow to

accommodate M-commerce applications. The

novelty of our proposal is that it uses an ontological

context model to provide personal and

environmental contextual information and supports

the composition of context-aware services. As a

consequence, it not only utilises existing web service

and service discovery protocol, but also employs

mobile agents to achieve flexible network roaming

for interactive services.

Even though we have completed a few critical

experiments, the whole research project is still in its

very early stage. Our next steps are to complete the

workflow management system and to integrate the

system with a mobile agent infrastructure. In

addition, there are some aspects that should be

further investigated. First, we shall study how to

model user behaviour though data mining and

reasoning, and how to predict the user’s actions

based on various profiles. Secondly, we will

redesign the mobile portal in order to manage the

limited computational resources of handhold devices

and provide a user friendly interface. Thirdly, we

will investigate the development of M-commerce

agents with more intelligent decision-making and

learning capabilities in the context of automated

business transaction.

ACKNOWLEDGEMENTS

The authors would like to thanks to the Natural

Science Foundation of P. R. of China (90912003,

60773108 and 90812001) and the Natural Science

and Engineering Council of Canada for supporting

this research.

REFERENCES

Ardissono, L. Furnari, R., Goy, A., Petrone, G., Segnan. M,

2007. Context-Aware Workflow Management, In

ICWE’07, Seventh International Conference on Web

Engineering, Como, Italy pp. 47-52.

Bădică, C., Ganzha, M., Paprzycki, M., 2005. Mobile

Agents in a Multi-Agent E-Commerce System, In

Proc. Of SYNASC, Timisoara: IEEE Computer Society

Press, pp. 207-214.

Bai, L., Chou, D. C., Yen, D. C., Lin, B., 2005. Mobile

commerce: its market analyses, International Journal

of Mobile Communications, 3(1), 66 - 81

Chakraborty, D., Joshi, A., Finin, T., Yesha, Y., 2005.

Service Composition for Mobile Environments,

Journal on Mobile Networking and Applications,

Special Issue on Mobile Services, 10, pp. 435-451.

Dey, A. K. 2001. Understanding and Using Context.

Personal and Ubiquitous Computing, Personal and

Ubiquitous Computing Journal, 5(1).pp. 4-7.

Dey, A. K., Abowd, G. D., 1999. Towards a Better

Understanding of Context and Context-Awareness.

Technical Report GIT-GVU-99-22, Georgia Institute

of Technology, College of Computing, Atlanta,

Georgia, USA.

Hagen, L., Breugst, M., Magedanz, T., 1998. Impacts of

mobile agent technology on mobile communication

system evolution, IEEE Personal Communication, 5,

pp. 56-69.

Kotz， D., Gray, R. S., 1999. Mobile Agents and the

Future of the Internet, ACM Operating Systems

Review, 33(3), pp. 7-13.

Kowalczyk, R., Braun, P., Frankczyk, B., Speck, A., 2003.

Deploying Mobile and Intelligent Agents in

Interconnected E-marketplaces, Journal of Integrated

Design and Process Science, 7(3), 109-123.

Laudon, K., Traver, C., 2006. E-commerce: Business,

Technology, Society, Prentice Hall. Boston, 3

edition.

Luo, N., Yan, J., Liu, M., Yang, S., 2006. Towards

Context-Aware Composition of Web Services, In Fifth

ICE-B 2010 - International Conference on e-Business

114

International Conference on Grid and Cooperative

Computing, Washington, DC, USA, pp. 494–499.

Mihailescu, P., Binder, W., Kendall, E, 2002. MAE: a

mobile agent platform for building wireless

m-commerce applications, In 8th ECOOP Workshop

on Mobile Object Systems: Agent Applications and

New Frontiers, Malaga, Spain.

Milanovic, N., Malek, M. 2004. Current Solutions for Web

Service Composition, IEEE Internet Computing, 8, pp.

51-59.

Ngai, E. W. T., Gunasekaran, A., 2007. A review for

mobile commerce research and applications, Decision

Support Systems, 43(1) pp.3-15.

Payne, T., Sycara, K., 2002. DAML-S: Web Service

Description for the Semantic Web, Lecture Notes in

Computer Science, 2342(2002), pp. 348-365.

Qiang, W., Hin, H. K. P., 2002. Agent-Based System for

Mobile Commerce, In IPDPS’02, 16th International

Parallel and Distributed Processing Symposium,

Florida, pp. 56-60.

Sadeh, N., 2002. M-commerce: Technologies, Services and

Business Models, John Wiley & Sons. New York, 1

edition.

Sandholm, T. and Huai, Q., 2000. Nomad: Moblie Agent

System for an Internet-Based Auction House, IEEE

Internet Computing, 4(2), pp. 80-86.

Uschold, M., Gruninger, M., 1996. Ontologies : Principles

, methods , and applications, Knowledge Engineering

Review, 11 (2) pp.93～155.

X. Li, 2006. On the Implementation of IMAGO System,

International Journal of Computer Science and

Network Security, Vol. 6, No. 2A, pp. 107-118

INTEGRATING MOBILE AGENT AND CONTEXT-AWARE WORKFLOW ANALYSIS FOR M-COMMERCE

APPLICATIONS

115