An intelligent product-information presentation in E-commerce S.S. - - PDF document

An intelligent product-information presentation in E-commerce

S.S. Manvi, P. Venkataram *

Protocol Engineering and Technology (PET)-UNIT, Electrical Communication Engineering Department, Indian Institute of Science, Bangalore 560012, India Received 2 May 2003; received in revised form 21 May 2004; accepted 12 January 2005 Available online 31 May 2005

Abstract Electronic commerce (E-commerce) web-sites must be equipped with multimedia presentations for effective market- ing of their products. Providing required product-information to a genuine buyer is a complex task in the present day web-based service environments. In this paper, we propose a distributed proxy based electronic shopping model, which is intelligent enough to study the customer behavior and plan the presentations accordingly by using a flexible multi- media synchronization model. The multimedia synchronization model is located at the proxy. The model triggers one of the three synchronization mechanisms, point, real-time continuous or adaptive synchronization based on the customer buying probability. The syn- chronization scheme employs a set of static and mobile agents: to estimate the network delays, to compute the skew, to monitor the loss and estimate the playout times of the presentation units of product-information. We simulated the electronic shopping model and the synchronization model to evaluate their operation effectiveness in several network scenarios. The benefits of scheme are: intelligent planning of product-information presentations, asynchronous delay estimation, flexibility and adaptability. 2005 Published by Elsevier B.V.

Keywords: E-commerce; Agents; Multimedia; Synchronization; Customer behavior

• 1. Introduction

With increasing number of Internet users and the rapid growth of networking technologies, Elec- tronic commerce (E-commerce) is perceived as one

• f the killer applications of the computer and com-

munication technologies. E-commerce can be de- fined as ‘‘the buying and selling of information,

1567-4223/$ - see front matter 2005 Published by Elsevier B.V. doi:10.1016/j.elerap.2005.01.001

* Corresponding author. Tel.: +91 803340855; fax: +91

803347991. E-mail addresses: sunil@protocol.ece.iisc.ernet.in (S.S. Manvi), pallapa@ece.iisc.ernet.in (P. Venkataram). Electronic Commerce Research and Applications 4 (2005) 220–239 www.elsevier.com/locate/ecra

products, and services via Computer networks’’. It has changed the way that sellers distribute their products and services to customers. Several ap- proaches to online shopping over a computer net- work are: Visiting a vendor site and search/browse for a product. Compilation of voluntary user ratings and reviews of vendor products (Recommender system). Comparison-shopping for finding products under best terms among the vendors. The works given in [1–3] uses buyer mobile agents that are dispatched to vendor sites, where they negotiate orders and deliveries, and returns to the buyer with their best deals for approval. Some

• f

the URLs listing

• f

agent based E-commerce sites are listed in [4]. Secure agent fabrication, evolution, and roaming architecture is proposed in [5] that manages and serves agents in E-commerce. Working of recommender systems in E-commerce and some of the related work are presented in [6]. Some of the works are based on comparison-shopping agents, which query several sites to gather product-information and provide a virtual market place for the customers [7–9]. The work given in [10] provides private labeled Voice-Over-IP calling, callback and web collabo- ration solutions for Internet commerce. An intelli- gent shopping architecture is proposed in [11], which learns user personal preferences and auton-

• mously shops on their behalf while protecting

their privacy. The mission of E-commerce is to help shoppers zero in on a product they want to buy and deter- mine which vendor they want to buy it from based

• n price, reputation, product availability and ser-
• vice. Shopping web-sites should be designed in

an attractive way to convert browsers into buyers: it should be user friendly and easy to navigate with proper product-information presentation [12–14]. A customer should be provided with sufficient information about the product to make buying decisions especially in case of high budget products such as automobiles, biomedical equipments, com- mercial softwares, etc. Multimedia presentations are effective market- ing tools that can empower any E-commerce site whether it is selling products or services. The work given in [15] describes about the use of 3D virtual

• bjects for entertaining and managing product cat-

egories, as well as maximize profitability of prod- uct categories. A virtual market place by using 3D objects representing the buyer agents, seller agents, market and products is presented in [31] that represents future trading environments. Multimedia stream presentations (audio, video, and images) increase the effectiveness of product- information presentations which may enhance the buyers confidence in purchasing. We made a survey of the E-commerce users (a sample size

• f 400) to get statistics of persons interested in

buying over the Internet and the way they wanted the product-information presentation to be. It was found that 98% of the users opted for multimedia presentations (some kind of multimedia demos to get the feeling of the product) rather than textual based presentations to zero in on the product purchasing. Multimedia presentations pose a problem of in- creased downloading time, where customers are made to wait for longer time. This problem can be eliminated by using streaming medias. Streaming technology reduces the playout latency, since there is no need to wait for complete downloading of a file: hence presentation can be started immediately after several parts of file are downloaded. SMIL (synchronized multimedia Internet language) is used for multimedia presentations in E-commerce, which performs only timeline-synchronized presen- tations [16]. Multimedia product-information presentations in E-commerce require intelligent and flexible pre- sentation, synchronization mechanisms to opti- mally utilize the network resources such as bandwidth, buffers, etc., as well as persuade the customers to purchase the products online. The behavior of the customers differ from one person to another. Some of the users of E-commerce sites may not be genuine buyers (they may be just surf- ing the web), hence it is not required to waste the costly resources (either network or server re- sources) for them by providing detailed multime- dia presentations

• f

the product-information.

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 221

Thus, there is a need to study the customer behavior and present the product-information presentations accordingly so as to maximize the profitability of E-commerce sites, and also manage the network and server resources efficiently. The customer behavior can be used to: distinguish between the genuine buyers and web surfers, judge probability

• f product purchases, and decide upon the type of

product-information presentations (detailed

• r

superficial). The multimedia presentations of a product- information facilitates the buyers to make the buy- ing decisions. The work given in [17] deals with modeling of customer behavior based on user vis- its and purchases, but, does not use any intelligent techniques to create product-information presen-

• tations. To the best of our knowledge, none of

the works deal with intelligent multimedia prod- uct-information presentations based on the cus- tomer buying behavior. This paper proposes a distributed proxy based based E-shopping model, that is intelligent enough to study the customer behavior and plan the product-information pre- sentations accordingly by employing different types of presentations based on customer purchase probability. 1.1. Need for synchronization Streams in E-commerce presentations such as video, audio, images, etc., are temporally related with each other. The process of maintaining these temporal relations during storage, transmission and presentation is called synchronization [18]. The problem of maintaining continuity within a single stream is referred as intra-stream or serial

• synchronization. Inter-stream or parallel synchroni-

zation deals with maintaining temporal relation- ships among the presentation units

• f

the multiple streams. Human perceptions about media synchroniza- tion and the presentation requirements for differ- ent applications are discussed in [19]. The maintenance of temporal relationships within or among the streams of the product presentations usually depends on following parameters [20]: end-to-end delay variations of presentation units, skew (interarrival time difference among the streams), delay jitters (interarrival time difference among the presentation units of the stream), end- system jitters, clock skew among servers and client, clock drift among servers and client, and change in generation and presentation rates. The works gi- ven in [21–23] describes the adaptive synchroniza- tion of periodic distributed multimedia streams based on delay estimations. The synchronization techniques given in [23,24] describes about the real-time, point and adaptive synchronization for hypermedia presentations. 1.2. Proposed work In this paper, we propose a distributed proxy based electronic shopping (E-shopping) model, which is intelligent enough to study the customer behavior and plan the presentations according to customer buying probability. The presentations are made by using a flexible multimedia streams synchronization model. The decision about pro- viding presentation depends on resources avail- ability and the customers past history

• f

resources accessed. The multimedia information about the product is distributed stream-wise among several servers of each vendor. Proxy fetches the streams from the distributed servers

• f a vendor selected by the customer and delivers

a synchronized presentation to the user. To the best of our knowledge, we have not come across a synchronization method which is flexible enough to provide different types of syn- chronizations in general distributed multimedia applications and in E-commerce. The proposed synchronization approach in E-commerce for product-information presentation is flexible and adaptable, since it provides three types of synchro- nized presentations depending on the customer behavior (potential buyer or searching for the product or surfing the websites) detected by the

• proxy. The synchronization scheme uses agents

to estimate network delays, clock differences, and playout times, which enables easy encoding of intelligence and aggregate tasks into the agents to provide flexible and adaptable services. We propose three types of presentations with different synchronization requirements based on the customer behavior. The types of presentations

222 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

are: (1) random clips of a product; (2) some ran- dom clips and detailed specific features of a prod- uct; (3) detailed presentations of a product. The proposed synchronization model uses an adaptive synchronization for Type 3 presentation, point synchronization for Type 2 presentation a real- time continuous synchronization for Type 1 presentation. Point synchronization realizes that the start time

• f presentation units of the streams is synchro-

nized with a certain specified synchronization point among the streams. In real-time continuous synchronization, presentation units of the streams are synchronized with real-time axis; for example, the motion video with contents of 20 s should be presented for exactly 20 s. In adaptive synchroniza- tion, presentation times of units of the streams are readjusted at regular intervals to adapt to the net- work delay variations experienced by presentation units, and reduce packet losses caused due to late arrivals. 1.3. Organization of paper The paper is organized as follows. Section 2 ex- plains the proposed E-shopping system model. Proposed agent based product-information pre- sentation synchronization model is described in Section 3. Section 4 discusses the simulation re-

• sults. Finally, we conclude in Section 5.
• 2. E-shopping system model

We propose a proxy based E-shopping system model (see Fig. 1) that comprises of vendors stream servers (audio, video, images), zone wise proxy shopping servers and the customers. The customers could be using desktop computers, per- sonal digital assistants or laptop computers either connected through wireless or wired media. Ven- dor servers contain archives of product-informa- tion presentations, which will be sent to proxy servers depending on the type of presentation request. The purpose of using distributed stream servers is to facilitate adaptation to failures: if one of the server crashes or a link fails, at least other servers will cater services to the customers until the fault is

• rectified. Also storing presentations stream-wise

distributes the load across the network as com- pared to centralized server. Zone wise proxy serv- ers are used to speed up the E-shopping services to the customers and reduce the congestion at the shopping mall as compared to single proxy shop- ping server. In this Section, we describe the salient features of the model, types of multimedia presen- tations of the product and the functions of the proxy shopping server. 2.1. Salient features of the model Prepares virtual market place for customers using proxy shopping servers. Learns the customer behavior and computes the customer buying probability. Plans intelligent multimedia presentations for the customer based on customer buying behav- ior and invokes appropriate synchronization mechanisms for streams playout. Allow vendors to store multimedia presenta- tions stream-wise distributed in the Internet. Presentation access to a customer is decided based on the resource (bandwidth and buffers) availability and the customers past history of accessing. The proposed product-information presenta- tion approach is intelligent because of the follow- ing two features.

Vendors Stream 1 Servers Vendors Stream 2 Servers Vendors Stream 3 Servers Servers Shopping Proxy Zone Customers Customers

• Fig. 1. Proxy based E-shopping system model.

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 223

• 1. The E-shopping model uses the history informa-

tion of the customers (products purchased, resources accessed, etc.) to determine the poten- tial buyers and presents the product-information according to the customer buying behavior by

• ptimally utilizing the resources of the proxy

and the network.

• 2. The proxy uses a set of static and mobile agents to

adapt to the network delay and jitter variations, and prepares the presentation schedule as per the presentation category of the user. 2.2. Types of presentations Types of multimedia presentations of a product are as follows. Type 1: random clips of a detailed presentation

• f a product; real-time synchronization is used

for this kind of presentation. Type 2: some random clips and detailed specific important features presentation of a product; point synchronization is used for this kind of presentation. Type 3: detailed presentation of a product, which includes all features, demonstration of functioning of the product, and recordings of the feedback from certain individuals. Adaptive synchronization is used for this kind

• f

presentation. Now we illustrate all these kinds of presenta- tions and the synchronization schemes used in them by considering an example of purchasing a car over the Internet. 2.2.1. Type 1 presentation An example of Type 1 presentation is illustrated in Fig. 2(a). The cars side views, front and back views are displayed as random moving clips along with corresponding audio streams. To illustrate the real-time synchronization of Type 1 presentation consider the 10 units of data (presentation units or simply PUs) received at the proxy from the stream 1 and stream 2 servers for car information display as shown in Fig. 2(b). The proxy computes the starting playout time by using estimated delays to servers, clock differences and skew information. The initial playout time will be utilized to compute the playout times for all the received PUs as per the real-time synchronization

• requirements. That is, 10 PUs of each stream must

be displayed within the specified time (say 2 s) once the playout begins (see computed playtime in

• Fig. 2(b)).

Side View

• f car

Random clips Random clips Random clips Random clips

• f car

front view view of car Opposite side

• f car

Back view

(b) Synchronization model in Type1 (Real-time synchronization) (a) Type 1 presentation

Played 1 2 2 3 3 4 4 5 5 6 6 7 7 8 8 9 9 10 10 1 1 2 3 4 5 6 7 8 9 10 1 3 2 4 6 5 Drop 8th, 9th and 10th PUs 2 seconds arrival Stream1 Stream2 arrival PUs PUs Playtime Computed PUs initial Playout time 1 3 5 7

• Fig. 2. (a) Random clips presentation of a car, (b) real-time synchronized presentation of clips.

224 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

Due to random arrivals of the PUs because of random delays incurred in network, some data will miss presentation deadlines, but, they are main- tained in the buffers so as to play them later, if sub- sequent PUs do not arrive in time. From the Fig. 2(b), we observe that 7 clips are played whereas the 8th, 9th and 10th PUs of each stream are dis- carded due to their arrivals after 2 s. The clips 2nd to 7th are delayed for playout since the clips arrived after the presentation deadlines. In case, if a required clip is not available during its dead- line, the recent clip in the buffer will be played

• ut (for example, PU 2 is delayed, hence, PU 1 is

played in place of PU 2). 2.2.2. Type 2 presentation An example of Type 2 presentation is illustrated in Fig. 3(a). The cars side views, front and back views, and a detailed view of internal functioning

• f car are displayed as random moving clips along

with corresponding audio streams. To illustrate the point synchronization of Type 2 presentation consider the 10 PUs received at the proxy from the stream 1 and stream 2 servers for car information display as shown in Fig. 3(b). The proxy computes the initial playout time by using the estimated delays to servers, clock differ- ences and skew information. The proxy displays PUs if both stream 1 and stream 2 PUs are avail- able (for instance, PU 2 of both streams is dis- played when both are available at the proxy),

• therwise display the old PUs (most recent ones;

for example, 4th PU of the streams 1 and 2 are not available after 3rd PU is displayed, hence proxy plays 3rd PU again). From the Fig. 3(b), we observe that 10 clips are played by taking more than 2 s. In some cases, some PUs may not arrive at all, which will lead to indefinite waiting by the proxy. To overcome this problem, the proxy waits for a PU not more than 3 PU intervals after last PU is displayed. In case, if it does not arrive, waits for displaying the next PU. The clips 8th, 9th and 10th are played after 2 s. 2.2.3. Type 3 presentation An example of Type 3 presentation is illustrated in Fig. 4(a). The cars side views, front view, back view, internal details and the test ride are displayed

Side View

• f car

Random clips Random clips Random clips Random clips

• f car

front view view of car Opposite side

• f car

Back view

Played 1 2 3 4 5 6 7 1 2 3 4 5 6 1 1 1 2 3 3 4 5 5 6 7 8 8 9 10 2 seconds arrival Stream1 Stream2 arrival PUs PUs Playtime Computed PUs initial Playout time

Deatailed clips of func tioning of car

Only initial playout time is computed (a) Type 2 presentation (b) Synchronization model in Type 2 (Point synchronization) 8 9 10 7 8 9 10

• Fig. 3. (a) Random clips and some detailed presentation of a car, (b) point synchronized presentation of clips.

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 225

as detailed moving clips along with corresponding audio streams. In this presentation, duration of the presenta- tion are divided into n time intervals. The proxy computes the playout times of all the PUs of an interval by using the previous estimated delays and the time at which last PU is displayed in the previous interval, and plays the PUs by using point

• synchronization. For example, T(n) interval PU

playout times are computed based on estimated network delays in T(n 1) interval and the last PU displayed in T(n 1) interval (see Fig. 4(b)). This allows the proxy to cope up with the random delay variations of the network and preserve the semantics of the presentations. 2.3. Functions of proxy shopping server The proxy uses intelligent agent technology to prepare the comparison charts for the product and create pages for the product by visiting the vendor sites as well interacting with the

• ther proxies nearby, and also collects the infor-

mation about multimedia product-information presentation locations stream-wise. The proxy updates the history of every logged in user with respect to following data: number

• f visits made by the customer, number of pur-

chases, number of pages viewed for a product, number of times customer requested for presen- tation, and number of times customer presenta- tion request accepted. It allows the customers to browse the prod- ucts and provides synchronized multimedia presentations by fetching the streams from distributed vendor sites upon the customer request, and also places the order to vendors

• n behalf of customer (if customer wishes to

purchase). The proxy server computes the customer buying probability (BP) and the resource access success ratio (RASR) of a product as given in Algo- rithm 1 whenever a customer logs in. Algorithm 1: computation of product BP and RASR

• f a customer

Begin

• 1. Choose weight (w1) for previous purchases

and weight (w2) for buying interest such that w1 + w2 = 1.0.

• 2. Compute product purchase probability (PP):

PP = (number of times purchased during a visit)/(number of visits to site).

• 3. Compute buying interest probability (BIP):

BIP = number of viewed pages of product/ number of pages of product.

Side View

• f car

Detailed clips Detailed clips front view

• f car

Detailed clips Opposite side view of car Random clips Back view

• f car

Deatailed clips of func tioning of car Detailed clips

• f test rides

and car demo

(Adaptive synchronization)

T(n) interval T(1) interval

(b) Synchronization model in Type 3 Initial playout time

etimated network delays in T(n1) interval in interval Tn by using in interval T1 by using

(a) Type 3 presentation

estimated network delays in previous (0th) interval Uses point synchronization Uses point synchronization

• Fig. 4. (a) Detailed clips presentation of a car, (b) adaptive synchronized presentation of clips.

226 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

• 4. Compute

customer buying probability (BP) = PP*w1 + BIP*w2.

• 5. Compute the customer resource access success

ratio (RASR): RASR = (number of times suc- cessful in resource access)/(number of times resources requested for presentation).

• 6. Stop

End. Other factors which can be considered for com- puting BP are: customer family income, cus- tomer profession, customer employer, etc. The proxy server classifies customers under different categories of presentations and syn- chronization based on customer BP whenever a user logs into the E-commerce web site. The customers with higher values of BP will be given detailed presentations (Type 3) to facilitate them to zero in on product purchase. Random clips and some detailed special features of a product (Type 2) will be given to a person with average values of BP so that the customer would be able to decide about product pur-

• chase. The customers with low values of BP

are given random clips of a product (Type 1) since the customer is not sure of buying (i.e., may be just interested in surfing the net). See Algorithm 2 for classification of customers under presentation category. Algorithm 2: classification of customers under pre- sentation category Begin

• 1. If (customer BP > X) then classify customer

under Type 3 presentation.

• 2. If (customer BP > Y and customer BP 6 X)

then classify customer under Type 1 presentation.

• 3. If (customer BP 6 Y) then classify customer

under Type 2 presentation.

• 4. Stop.

End. where X and Y are values chosen by the proxy ser-

• ver. Typical values that can be considered for X

and Y are 0.75 and 0.5, respectively, which have been determined empirically by taking a sample size of 400 E-commerce users. The server takes a decision about providing the product-information presentations (if customer wishes to see a product presentation) based on the maximum number of customers that can be concurrently supported (max_cust_support) and the resources utilized by the presentation requesting users (Customer RASR) (see Algo- rithm 3). Algorithm 3: scheduling the presentations based on RASR Begin

• 1. Sort

the presentation requesting users according to RASR in ascending order.

• 2. If

(Requesting users 6 max_cust_support) then {Provide synchronized presentations to all the requesting users with respect to their classified synchronization presentation} Else {Select only the first max_cust_support customers for presentations from the sorted list, and present them with respect to their classified synchronization}.

• 3. Stop.

End. If user presentation request is not successful because of occupied resources by other custom- ers, the proxy server will compute the time when resources will be freed based on the completion time of presentations of occupying customers and notifies the customer to request for presen- tation at that time. The proxy server connects to vendor stream servers for streaming data transfer (data is based on type of presentation) once customer presentation request is granted. The synchroni- zation agency located at the proxy synchronizes the distributed streams according to type of pre- sentation planned and sends to customers for playout. The proxy server will try to optimize the resource usage among users by presenting only selected media streams based on customers buying behavior thereby allows the server to cater to more number of customers. The proxy server intelligently caches the prod- uct-information presentations, which are often requested by the customers.

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 227

The proxies prevent direct connection between vendors and customers, and thereby hides iden- tity of a customer from vendors. This will pre- vent the customer from receiving frequent mails from the vendor.

• 3. Proposed product-information presentation

technique The proposed flexible synchronization scheme employing different types of product-information presentations is deployed in the proxy shopping servers that uses three synchronization mecha- nisms: real-time continuous, point and adaptive

• synchronizations. The scheme employs a set of sta-

tic and mobile agents to assist synchronized pre- sentations. In this section, we give a brief description of agents and discuss the proposed product-information presentation framework. 3.1. Agents Agents are the autonomous programs situated within an environment (host/network), which sense the environment and acts upon it to achieve their goals. They have certain special properties which make them different from the standard programs such as mandatory and

• rthogonal properties. Mandatory properties of

the agents are: autonomy, reactive, proactive and temporally continuous. The orthogonal properties are: communicative, mobile, learning and believ- able [25]. Mobile agents migrate from one host to another host in a heterogeneous network, and executes at remote host until they accomplish their task. An agent platform situated at a host or an intermedi- ate node in the network supports agent mobility, persistence, security, interagent communication and agent execution. An agent platform comprises

• f agents, agent server, agent interpreter and trans-

port mechanisms for agents. Agent server is responsible for receiving mobile agents, sending it for execution by interpreter and inter-agent com-

• munication. Agent interpreter depends on the

agent script/language used. In general, there are several good reasons for using agents: they reduce network load; overcome latency; encapsulate protocols; execute asynchro- nously and autonomously; and adapt dynami-

• cally. Agent based schemes comprising of static
• r mobile agents offer several advantages, such

as, flexibility, adaptability, software reuse and maintainability, and thus improves overall perfor- mance of the system as compared to traditional methods [26]. 3.2. Playout system model In this section, we discuss a playout system model, which is considered for describing the prod- uct-information synchronization mechanisms. The model consists of distributed stream servers (vendor servers) and the client (proxy shopping server) as depicted in Fig. 5, in which the proxy c synchronizes the multiple streams S1 to Si arriving at it and sends for playout to the customers. The assumptions made in the model for deriving the playout times are as follows:

• 1. The presentation unit (PU) of the streams is

chosen as the unit perceivable by the human beings, for example, frame is a presentation unit

• f a video stream. The PUs of each stream is

labeled with sequence numbers, and they are presented in same sequence.

• 2. The network delay of a PU of each stream var-

ies stochastically. PUs does not arrive in order at the proxy shopping server.

C

Network S1....Si : Stream Servers S2 S3 Si S1 Vendor C: Client/proxy

cust

• mer
• Fig. 5. Playout system model.

228 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

• 3. Clock differences exist between the servers and

the proxy shopping server, and the clocks of the customers and the proxy are synchronized.

• 4. Rate of clock drifts is zero.
• 5. Stream servers and the proxy are equipped with

agent platform to support agent mobility. How- ever, in case of unavailability of an agent plat- form, the agents use traditional message exchange mechanisms to achieve its goals. The synchronization agency for product- information presentation is run at the proxy shopping server which is explained in the follow- ing section. 3.3. Product-information presentation framework The proposed synchronization agency frame- work for product-information presentation at the proxy comprises of the following components (see Fig. 6): user interface agent, synchronization agent, mobile agents, presentation table and syn- chronization profile. The functions of each of these components are given below: Synchronization profile: this profile stores the product presentation session information (sta- tic and dynamic) for the customer and facili- tates sharing of information among agents. The static information is set by the user inter- face agent whereas dynamic information is updated by either mobile agents or synchroni- zation agent. The static information stored are: application identification number, stream servers address, generation period, presentation period, sustainable losses, maximum allowed presentation skew and the length of interval (in terms of PUs) to reestimate the playout

• times. The dynamic information stored are:

clock differences between the proxy and vendor servers, estimated delays to vendor servers from the proxy, skew between the streams, data transfer start time of the streams, syn- chronization type and the monitored losses of each stream.

USER INTERFACE AGENT

SYNCHRONIZATION AGENT presenta tion table (seq no, time)

monitoring loss mobile agent creation start time, etc)

appid, syntype, synpar

application ID generation period presentation period sustainable losses server adressess

• no. streams

interval length estimated delays synchronization type clock differences

• bserved losses/stream

skew of streams server start time allowed skew playout start time (skew, playout estimation playout

Static information Dynamic information MOBILE AGENTS

(creates synchronization agent)

send PUs to customer

Customer PUs constru ction profile Product nization synchro presentation

• Fig. 6. A framework for product-information presentation.

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 229

Presentation table: this table maintains the refer- ence playout times of presentation units of the streams (sequence number and playout time) as computed by the synchronization agent of the customer. User interface agent: it is a static agent created at proxy server for the customer which col- lects application (product presentation) infor- mation such as application identification number (appid), stream servers address, syn- chronization type (syntype) required and the synchronization parameters (synpar). The syn- chronization parameters include sustainable and desired presentation rates, maximum allowed skew, and acceptable loss. It sets syn- type value based on the customer buying status, i.e., syntype = 2 for adaptive synchro- nized presentation, syntype = 1 for point synchronized presentations, and syntype = 0 for real-time synchronized presentations. It creates a synchronization agent and a prod- uct-information presentation synchronization profile for the customer at the proxy shopping server. Synchronization agent: this static agent per- forms three types of synchronizations based

• n the product presentation requirements. To

identify the presentation requirements, it uses a variable syntype (0 = real, 1 = point and 2 = adaptive). It creates a mobile agent for each stream to estimate the clock difference and the network delays on the path which runs from the vendor stream server to client (proxy), and computes the skew, stream data transfer start time and the playout start time. It computes reference playout times of all the PUs of each stream in case of real-time con- tinuous synchronization. In case of adaptive synchronization, it monitors losses of each stream, and periodically recomputes the play-

• ut times of all the PUs of each stream based
• n the recently estimated network delays and

the monitored losses of each stream. Synchro- nization agent updates the product-informa- tion presentation synchronization profile and the presentation table of the customer with newly estimated parameters and the playout times, respectively. The agent, to provide continuous playout

• f

the streams at the customer site, uses skew compensation mechanisms. Mobile agents: these agents estimate the clock differences and the network delays among the vendor servers and the proxy, and updates the synchronization profile of the customer. They are also used to inform the servers about the start time of data transfer. They can be programmed to monitor the band- width, loss and delay parameters at the inter- mediate nodes and perform parameter negotiation at the vendor servers, intermediate nodes and the proxy to facilitate better qual- ity presentation. The scheme does not require time stamping of PUs since the presentation timing of a PU is com- puted by the agency itself. Network delays are esti- mated in real time independent of PU arrivals enhancing the adaptive capabilities

• f

the synchronization. 3.4. Product-information streams synchronization Synchronization agent uses mobile agents to estimate the clock differences and the network de- lays on the paths, which run from the vendor stream servers to proxy, and computes the skew, stream data transfer start time and the playout start time. Here, we give a brief idea of how agents are used in achieving synchronized presentations

• f Types 1, 2 and 3. Algorithm 4 describes compu-

tation of the clock difference [27], network delays, skew, server data transfer start time, and the play-

• ut start time at the client side by considering a

session with s number of streams. The proxy will create a mobile agent for each stream. Mobile agent of the respective stream makes z round trips to stream server to estimate the network delays, where z 2 {1, 2, . . . , Z} and Z = maximum round trips. Algorithm 4: delay estimation and playout start time computation {To describe the mechanism consider a session with s streams}

230 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

Begin

• 1. User interface agent creates the synchroniza-

tion agent and updates the synchronization profile for product presentation with session information.

• 2. For St = 1 to s do

Begin Synchronization agent creates a mobile agent forvendorproduct-informationstream Stfor estimation of clock difference and network delays. Mobile agent of vendor product-information stream St makes Z round trips to vendor stream server St and records the clock differ- ence and the one way delay between the ven- dor stream server St and the proxy, computes the delay between the proxy and server, and updates the product presentation synchroni- zation profile. Endfor St;

• 3. For k = 1 to s do

Begin Synchronization agent computes the skew for the vendor product-information stream k and updates the product presentation syn- chronization profile. Synchronization agent computes the ven- dor product-information stream k trans- mission start time and sends a mobile agent to server k to convey start time infor- mation, and updates the product presenta- tion synchronization profile. Synchronization agent computes playout start time at the client of the vendor prod- uct-information stream k and updates the product presentation table. Endfor k;

• 4. Stop.

End. Algorithm 5 describes the real-time synchroniza- tion by using agents which uses algorithm 4 (Please refer Section 2 for description of real-time synchro- nization model). Synchronization agent sends the media units to the customer for playout that arrive within the estimated playout time. The agent, to

• vercome the playout gap problems caused due to

PU losses or late arrivals, uses skew compensation

• mechanisms. These mechanisms are: restricted

blocking (proxy sends the most recent frame to the customer for display to deal with the losses and delayed frames) and blocking (do not play any- thing) for video and audio streams, respectively. The late arrived PUs of each stream will be skipped. Algorithm 5: real-time synchronization {To describe the mechanism consider a session with s streams} Begin

• 1. Call Algorithm 4.
• 2. For St = 1 to s do {synchronization agent com-

putes the reference playout time of all the PUs

• f the vendor product-information stream St}.
• 3. For St = 1 to s do {synchronization agent

begins sending PUs of vendor product-infor- mation stream St to customer for playout by using skew compensation mechanisms}.

• 4. Stop.

End. Algorithm 6 describes point synchronization by using the agents which uses algorithm 4 (Please refer Section 2 for description of point synchronization model). Synchronization agent uses skew compen- sation mechanisms to compensate the delays. Algorithm 6: Point synchronization {To describe the mechanism consider a session with s streams} Begin

• 1. Call Algorithm 4.
• 2. For St = 1 to s do {synchronization agent

begins sending PUs of the vendor product- information stream St to customer for playout by using skew compensation mechanisms}.

• 3. Stop.

End. Adaptive synchronization is described in Algo- rithm 7 (Please refer Section 2 for description of adaptive synchronization). Synchronization agent sends a mobile agent of each stream periodically

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 231

(at regular time intervals) for Z/2 trips to estimate the network delays over the path from vendor ser- ver to proxy. These estimated delays and the ob- served PU losses are used to readjust the reference playout times of the PUs, The gaps cre- ated due to readjustment of reference playout times are compensated by using skew compensation

• mechanisms. The reference playout times are com-

puted for every k presentation units (k is the length

• f time interval for delay adaptation).

Algorithm 7: adaptive synchronization {To describe the mechanism consider a session with s streams, T time intervals, k PUs in each time interval} Begin

• 1. Call Algorithm 4.
• 2. For t = 1 to T do

Begin If (t == 1) {synchronization agent computes the playout times of k PUs of vendor prod- uct-information streams for interval t}. If ((losses in the interval t 1 >acceptable loss) and t > 1) {synchronization agent recomputes the playout times of k PUs of all the vendor product-information streams in the interval t by using the estimated delays and the time at which the last PU is displayed in the previous interval}. Else {compute the playout times of k PUs of the vendor prod- uct-information streams in the interval t (for t > 1) by using the time at which last PU is displayed and the old estimated delays}. For St = 1 to s do Begin – Synchronization agent starts sending PUs of vendor product-information stream St to customer for playout in time interval t by using skew com- pensation mechanisms. – Synchronization agent monitors the losses of vendor product-information stream St in time interval t and updates the product presentation synchronization profile. – Synchronization agent sends a mobile agent for Z/2 trips to estimate the net- work delays between the vendor prod- uct-information stream server St and the proxy, and updates the product presentation synchronization profile. Endfor St; Endfor t;

• 3. Stop.

End. Eventhough multimedia stream presentations increase effectiveness of product-information pre- sentation and also enhance the purchasing proba- bility, but, it has some overheads to provide synchronized presentations. The overheads are: buffers to smooth the delay jitters, bandwidth re- quired for mobile agents agents to estimate the net- work delays, an agent platform at the nodes, and maintenance of synchronization profiles. Never- theless an agent sent across a network can perform flexible and adaptable operations. Agent oriented programming facilitates component-based soft- ware engineering (CBSE) which is needed in to- days software development of web-based systems [30]. In future, there will be enormous number of agents (agents are next-generation components) which have to coordinate with each other to pro- vide better virtual market places, multimedia infor- mation searching, retrieval, and communication services, once the agent platform is standardized. Some of the benefits of using the agents in prod- uct-information presentation are: The agents allow learning capabilities to be incorporated in a natural way to support prod- uct information changes, changes in stream servers, network delay predictions, playout decision making depending on shopping server and the customer loads. Facilitates software reuse because of autono- mous operation of agents, i.e., an agent devel-

• ped by one shopping server can be reused by
• ther shopping servers by making slight modifi-

cations if necessary. Ease of software maintenance due to decom- posed tasks, i.e., since agents are autonomous

232 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

and developed based on independent modular

• approach. Debugging and updating of the E-

shopping and communication software will be easier. Flexibility in delay estimation policies by chang- ing mobile agent code facilitates personalizing

• f services by the proxy, i.e., estimation policies

may be coded to depend on the time of day, nat- ure of earlier predicted traffic at that time and successive time intervals. A delay estimation agent can be coded to per- form aggregate tasks such as bandwidth alloca- tion and monitoring [28,29], QoS (Quality of Service) negotiation and renegotiation, loss detection, changes in prices, special discounts

• f the products of the vendor, etc.

Provides faster way of protocol development and facilitates customized protocol implementa- tion. Agents encapsulate protocols, which allows speeding up of protocol implementation since there is no need to wait for lengthy stan- dardization process.

• 4. Simulation

We have carried out the simulation of the cus- tomers behavior and the product information pre- sentations by using the synchronization model for E-shopping services on wired and wireless net-

• works. In this simulation, we have considered sev-

eral products and their information in the form of multimedia streams available at several servers. We have also considered the case of one products information distributed among the many servers based on the information level (surface level, brief note, detailed level, etc.). We tested the designed product-information presentation model at the proxy for several customers needing multimedia data from three stream servers of a vendor. 4.1. Simulation model Several assumptions are made in the simulation model which is as given below. The number of visits (mv) made by a customer to proxy server are randomly distributed between 1 and maxv, where maxv is the maxi- mum visits. The number of times the customer purchased (pr) among the visits is randomly distributed between 1 and mv. The number of pages viewed for a product by the customer is randomly distributed between 1 and mp, where mp is the maximum pages of a product. We assigned equal weightage to customer pur- chasing and buying interest probability. Customer resource access ratio is randomly dis- tributed between 0 and 1. Maximum number of customers supported by a proxy for presentations are max_cus_sup. Probability of a logged in user asking for product presentation is Bernoulli distributed (Pup). A customer is assumed to purchase if a random variable generated between the range [BP, 1.0] is greater than 0.75, where BP is the buying prob- ability of a customer. Network delays are normally distributed with mean l and variance r2. The delay rises at every regular interval (interval in terms of number of PUs) and is Bernoulli distributed with probabil- ity, P. Increase in mean delay of PUs due to P is dr*l, where dr is the percentage delay rising factor. Increase in delay variance due to P is V*(l + dr*l), where V is the percentage rising factor. Application specifies: the number of PUs (N), acceptable/sustainable losses, PU generation and presentation period, and the skew tolerance in terms of PUs. 4.2. Simulation procedure The input values considered for simulation are: maxv = 20; mp = 10; max_cus_sup = 100; w1 = w2 = 0.5; Pup is considered as 0.2, 0.5 and 1.0; number of customers at the proxy are varied from 50 to 500; number of PUs to be presented (N) = 20,000, 12,000 and 6000 PUs for Type 3, Type 2 and Type 1 presentations, respectively; number of streams = 3; network delays (in milli- seconds) of stream 1, stream 2 and stream 3 are

S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 233

normally distributed with l1 = 500, l2 = 400, l3 = 450, r2

1 ¼ 100

and r2

2 ¼ 100; r2 3 ¼ 100 ms;

Probability of delay rise P is varied from 0 to 1.0; interval of PU length (k) for considering delay rise = 200 PUs; rising factor dr is randomly distrib- uted between 1% and 100% and V is randomly dis- tributed between 1% and 20%; presentation and generation rate = 80 ms; acceptable loss = 10% for all streams; and maximum allowed skew for presentation = 100 ms. The simulation procedure is as follows.

• 1. Simulated the E-shopping model to learn the

customer behavior and plan the product presentations.

• 2. Point synchronization mechanism was simu-

lated for one set of the customers to find

• ut the optimal number of trips required by

the mobile agents to estimate the network delays.

• 3. Real-time synchronization mechanism is simu-

lated for one group of the customers employ- ing

• ptimal

number

• f

mobile agents as computed

• evaluate

synchronization loss and buffering delays in case

• f

delay fluctuations.

• 4. Adaptive synchronization mechanism was sim-

ulated for one group of the customers to ver- ify the synchronization loss and mean buffering delays in case of delay fluc- tuations. The performance parameters evaluated in the simulation are as follows: Number

• f

customers in a synchronization category: it is defined as the number of cus- tomers at a proxy classified under different presentation categories based

• n

buying probability. Number of customers given resource access: it is defined as the number of customers success- ful in accessing the resources for presentation among the total number of customers request- ing for presentation at a given instant of time. Synchronization overheads: it is defined as the bandwidth required by the mobile agents to estimate network delays. Number of customers purchased: it is defined as the total number of customers purchased a product at the shopping proxy. Stream PU loss: it is defined as the percentage

• f PU losses in a stream.

Synchronization loss: it is defined as the percent- age of PU loss of either of the streams in a pre- sentation period. Mean buffering delay: it is defined as the mean waiting time of presented PUs of a stream in the receiver (proxy) buffer. 4.3. Results We observe that maximum number of users fall under real-time synchronized presentation cate- gory (see Fig. 7 with the considered simulation in- puts). We have also noticed that by increasing the lower limit of pr and mv, number of customers fall- ing under point and adaptive synchronized presen- tation categories increase since the customer buying probability increases. It is observed that number of customers given resource access in- creases with increase in probability of logged in customer asking for presentation (Pup), and can cater presentations for maximum of 100 customers (max_cus_sup = 100) which is the upper limit on proxy for servicing multiple customers (see Fig. 8).

50 100 150 200 250 100 150 200 250 300 350 400 450 500 Number of customers in a Synchronization category Number of customer arrivals Customer classification according to synchronization category

Real time sync. Point sync. Adaptive sync

• Fig. 7. Number of customers classified with respect to syn-

chronization category based on customer buying probability vs. number of customer arrivals at proxy. 234 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

We observe from Fig. 9 that the synchroniza- tion overheads increase with increase in number

• f customer arrivals and the Pup. The number of

customers purchasing a product will increase with the increasing number of customer arrivals and the maximum number of customers purchased are 89

• ut of 100 maximum product viewers (see Fig. 10).

The plots in Figs. 11 and 12 are used to deter- mine the optimal number of trips required by a mobile agent to estimate the delays without con- sidering delay fluctuations (P = 0). We observe that losses and the buffering delays almost remain stable for trips = 20. So, we initially, fixed the mo- bile agents trips to be 20 for real-time and adaptive synchronization. We observe (Figs. 13 and 14) that the PU losses increase with delay fluctuations (by varying P), and buffering delays of PU of the streams reduce with increase in losses and P. Adaptive synchronization results are shown in

• Figs. 15 and 16. We can observe that the losses

are maintained within the acceptable limit (10%), and the buffering delays are slightly higher for all

10 20 30 40 50 60 70 80 90 100 100 150 200 250 300 350 400 450 500 Number of customers given resource access Number of customers logged Number of customers given resource access Pup = 0.2. Pup=0.5 Pup=1.0

• Fig. 8. Number of customers given resource access for different

values of Pup vs. number of customers logged in at proxy.

0.2 0.4 0.6 0.8 1 1.2 1.4 50 100 150 200 250 300 350 400 450 500 Synchronization overheads (Mbps) Number of customer arrivals Synchronization overheads Vs. Customer arrivals Pup=0.2 Pup=1.0

• Fig. 9. Synchronization overheads vs. number of customer

arrivals at proxy.

10 20 30 40 50 60 70 80 90 50 100 150 200 250 300 350 400 450 500 Number of customers purchased Customer arrivals Pup=0.2 Pup=1.0 Number of customers purchased Vs. Customer arrivals

• Fig. 10. Number of customers purchased vs. number of

customers arrivals at proxy.

stream 3 2 4 6 8 10 12 14 16 5 10 15 20 25 30 35 40 45 50 PU loss (%) Number of trips made by mobile agents PU loss (%) Vs. Number of trips made by mobile agents stream 1 stream 2 Stream 1 or 2

• Fig. 11. PUs loss (%) vs. number of trips made by mobile

agents with P = 0 (without injecting delay rise). S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 235

the streams as compared to other plots (see Figs. 12 and 14). Due to continuous adaptation to changes in delays, PUs will be buffered for more time to avoid loss. We have also done experiments by changing the generation rates and delay rising factors to exten- sively validate the synchronization scheme: it is no- ticed that the nature of results remain same, but the synchronization losses increase and mean buf- fering delay decreases with increase in delay rising factor. 4.4. Benefits of using agents We observed the following benefits of using agents in product-information presentation in E-commerce while simulating the proposed framework. Intelligent planning of presentations: the E-shop- ping model uses the history information of the cus- tomer (products purchased, resources accessed, etc.) to determine the potential buyers and pre- sents the product-information according to the

stream 3 1 2 3 4 5 6 7 8 9 10 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 PU loss (%) Probability of delay rise, P PU loss (%) Vs. Probability of delay rise stream 1 stream 2 Stream 1 or 2

• Fig. 15. PU loss (%) vs. probability of delay rise (P) for

adaptive synchronization.

stream 3 10 20 30 40 50 60 70 80 0.2 0.4 0.6 0.8 1 PU loss (%) Probability of delay rise, P PU loss (%) Vs. Probability of delay rise stream 2 Stream 1 or 2 stream 1

• Fig. 13. PU loss (%) vs. probability of delay rise (P) for real-

time synchronization.

stream 3 100 120 140 160 180 200 220 240 260 0.2 0.4 0.6 0.8 1 Mean buffering delay Probability of delay rise Mean buffering delay (ms) Vs. Probability of delay rise stream 1 stream 2

• Fig. 14. Mean buffering delay for a PU vs. probability of delay

rise (P) for real-time synchronization.

stream 3 160 180 200 220 240 260 280 10 15 20 25 30 35 40 45 50 Mean buffering delay Number of trips made by mobile agents Mean buffering delay (ms) Vs. Number of trips made by mobile agents stream 1 stream 2

• Fig. 12. Mean buffering delay for a PU vs. number of trips

made by mobile agents with P = 0 (without injecting delay rise). 236 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

customer buying behavior by optimally utilizing the resources of the proxy and the network. Asynchronous delay estimation: the agents

• perate asynchronously, i.e., they do not need

permanent connections between a source and des-

• tination. This facilitates the source to carry on

some other tasks instead of waiting for the sent

• agent. Thus, an agent sent across the network will

travel on its own and report back its round trip delay to the source. This feature is especially more useful in case of wireless networks where frequent disconnections are possible. Flexibility: flexibility in delay estimation poli- cies, product-information presentation planning, customer behavior prediction is possible in pro- posed framework. Delay estimation policy can be changed by changing mobile agent code which al- lows user personalize his services, i.e., estimation policies may be coded to depend on the time of day, nature of earlier predicted traffic at that time and successive time intervals. Product-information presentation planning such as type of presenta- tions, number of clips required in each category

• f presentations, resolution, etc., can be changed

by the E-shopping server from time to time based

• n some criteria. Code to predict customer behav-

ior can be easily changed by employing different weights to the resource access and the type of items purchased. Adaptability: the mobile agents enable the proxy to adapt to the estimated network delays and prepare the presentation schedule as per the presentation category of the user. The E-shopping server adapts to the changes in network and server resources by planning different types of presenta- tions according to the user behavior. Support to CBSE: the agent based software development facilitates component based software engineering (CBSE) considerations such as main- tainability of agent codes, reusability of agent code, flexibility and adaptability of services.

• 5. Conclusions

In this paper, we proposed a distributed proxy based electronic shopping model, which is intelli- gent enough to study the customer behavior (whether customer is a browser or purchaser) and plan the product-information presentations accordingly by using an agent based synchroniza- tion model. The model takes a decision about the type of presentation (and synchronization scheme) to be employed for a customer depending on re- sources availability and the customers past history

• f resources accessed. It triggers one of the three

synchronization mechanisms, real-time continu-

• us, point or adaptive synchronization based on

the customer buying probability. A purchaser is gi- ven enough information about the product to make a decision. The multimedia synchronization model is run at the proxy. Adaptive synchronized presentations facilitates a customer with high buy- ing probability to thoroughly look at the product to improve the purchase confidence. We simulated the electronic shopping model to evaluate the operation effectiveness in several net- work scenarios. The benefits of the scheme are: (1) intelligent product-information presentation planning based on the customer buying probabil- ity that aids in proper utilization of network and server resources, (2) distributed proxies speed up the shopping services by distributing the data stream-wise and also facilitate to overcome multi- media server failures, (3) flexible and adaptable synchronization services by allowing the user to encode the delay estimation policies and choose

stream 3 240 260 280 300 320 340 360 380 400 420 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Mean buffering delay Probability of delay rise Mean buffering delay (ms) Vs. Probability of delay rise stream 1 stream 2

• Fig. 16. Mean buffering delay for a PU vs. probability of delay

rise (P) for adaptive synchronization. S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 237

proper synchronization mechanism to view the material based on the customer behavior, (4) sup- ports component based software engineering as- pects such as software reuse, maintainability and customizability. Even though agent technology offers more flex- ible adaptation services, several problems have to be still researched and resolved in mobile agent systems implementation, such as, security to hosts and agents, creation of agents and agent coding languages. The proposed presentation framework for E-commerce can also be applied to other multime- dia applications, (both in wireless and wired net- work) by making minor modifications to the

• framework. For example, it can be applied to an
• nline Internet-based education systems where dif-

ferent types of synchronizations are required for viewing the education material (like random clip viewing, detailed viewing, etc.). The adaptive syn- chronization and real-time synchronization of the framework can be used for video-conferencing application

• f

Internet, i.e., application can dynamically employ either real-time synchroniza- tion (whenever the network load is low) or adap- tive synchronization (whenever the network load is high or abruptly increasing). Acknowledgments We thank the anonymous reviewers for their valuable suggestions that helped us to improve the presentation quality of the paper. References

[1] P. Dasgupta, N. Narasimhan, L.E. Moser, P.M. Melliar- Smith, MAgNET: mobile agents for networked electronic trading, IEEE Transactions on Knowledge and Data Engineering 11 (4) (1999) 509–525. [2] A.R. Greenwald, J.O. Kephart, Shopbots and pricebots, in: Proceedings of the International Joint Conference on Artificial Intelligence (IJCAI), 1999, pp. 1–23. [3] P. Maes, R.H. Guttman, A. Moukas, Agents that buy and sell, Communications of ACM 42 (3) (1999) 81–91. [4] W. Mougayar. Survey on the future of agent-based commerce on the web. Available from: <http://www. cyberm.com/art2.htm>. [5] S. Guan, F. Zhu, Agent fabrication and its implementation for agent based E-commerce, Journal of Information Technology and Decision Making 1 (3) (2002) 473–489. [6] J.B. Schafer, J.A. Konstan, J. Riedl. Recommender systems in E-commerce, in: Proceedings of the ACM Conference on Electronic Commerce (EC-99), 1999, pp. 158–166. [7] R.B. Doorenbos, O. Etzioni, D.S Weld, A scalable comparison-shopping agent for the World Wide Web, in: Proceedings of the Ist International Conference Autono- mous Agents, 1997, pp. 39–48. [8] BANDELLO: Real time comparison shopping system. Available from: <http://www.myGoldDigger.com>. [9] Mysimon Internet Commerce. Available from: <http:// www.mySimon.com>. [10] Surf and Call E-commerce Solution. Available from: <http://www.surfandcall.net/main/>. [11] F. Menczer, W. Nick Street, A.E. Monge, Adaptive assistants for customized E-shopping, IEEE Intelligent Systems 17 (6) (2002) 12–20. [12] N.J. Kublius. Designing an E-commerce site for users. ACM cross roads student magazine. Available from: <http://www.acm.org/crossroads/xrds7-1/ecuser.html>. [13] L.J. Najjar, E-commerce user interface design for the Web, in: Proceedings of the HCI International, 2001, pp. 843– 847. [14] K. Kang, Effectiveness of graphical components in web site E-commerce application – a cultural perspective, The Electronic Journal on Information Systems in Developing Countries 7 (2) (2001) 1–6. [15] R.R. Burke, Virtual shopping: breakthrough in marketing research, Harvard Business Review 74 (2) (1996) 120–131. [16] S.S.Y. Shim, J.Z. Gao, Y. Wang, Multimedia presentations in E-commerce, in: Proceedings of the International Workshop on Advanced Issues of E-commerce and Web- based Information Systems (WECWIS), 2000. [17] Wendy W. Moe, Peter S. Fader, Dynamic conversion behavior at E-commerce sites. Available from: <http:// www-marketing.wharton.upenn.edu/ideas/pdf/Fader/ Fader>. [18] G. Blakowski, R. Steinmetz, Media synchronization survey: reference model, specification, and case studies, IEEE Selected Areas in Communications (JSAC) 14 (1) (1996) 5– 35. [19] R. Steinmetz, Human perception of jitter and media synchronization, IEEE JSAC 14 (1) (1996) 61–72. [20] T.V. Johnson, A. Zhang, Dynamic playout scheduling algorithms for continuous multimedia streams, Multimedia Systems 7 (1) (1999) 312–325. [21] N. Shivkumar, C.J. Sreenan, B. Narendra, P. Agarwal, The concord algorithm for synchronization of networked multimedia streams, in: Proceedings of the International Conference on Multimedia Computing Systems (ICMCS), 1995. [22] K. Rothermel, T. Helbig, An adaptive protocol for synchronizing media streams, Multimedia Systems 5 (5) (1997) 324–336. 238 S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239

[23] K. Fujikawa, S. Shimojo, T. Matsuura, S. Nishio, H. Miyahara, The synchronization mechanisms of multimedia information in the distributed hypermedia system ‘‘Har- mony’’, in: Proceedings of the International Conference on Multimedia Modeling, 1993, pp. 275–289. [24] S.S. Manvi, P. Venkataram, A multimedia synchronization model for online Internet based education systems using agents, Journal of Computer Science and Informatics 33 (2) (2003) 31–41. [25] S.S. Manvi, P. Venkataram, Applications of agent tech- nology in communications: a review, Computer Commu- nications Journal 27 (15) (2004) 1493–1508. [26] N.R. Jennings, An agent-based approach for building complex software systems, Communications of ACM 44 (4) (2001) 35–41. [27] W. Jiang, H. Schulzrine, QoS measurement of Internet real-time multimedia Services. Technical report CUCS- 015-99, Columbia University, 1999. [28] S.S. Manvi, P. Venkataram, Mobile agent-based online bandwidth allocation schemein multimediacommunication, in: Proceedings of the IEEEGlobecom, 2001, pp. 2622–2626. [29] S.S. Manvi, P. Venkataram, QOS management by mobile agents in multimedia communication, in: Proceedings of the Database and Expert Systems (DEXA), 2000, pp. 407–411. [30] M.L. Griss, G. Pour, Accelerating development with agent components, IEEE Computer Magazine 34 (5) (2001) 37–43. [31] Peter Dalmaris, Paul Bogg, Future trading environments, in: Proceedings of the Workshop on Smart E-business, Indian International Conference on Artificial Intelligence, Hyderabad, India, 2003, pp. 51–64. S.S. Manvi, P. Venkataram / Electronic Commerce Research and Applications 4 (2005) 220–239 239