CHANGE DETECTION AND MAINTENANCE

OF AN XML WEB WAREHOUSE

Ching-Ming Chao

Department of Computer and Information Science, Soochow University, Taipei, Taiwan, R.O.C.

Keywords: Web warehouse, XML, Ch

ange detection, Warehouse maintenance, Mobile agent

Abstract: The World Wide Web is a popular broadcast medium that contains a huge amount of information. The web

warehouse is an efficient and effective means to facilitate utilization of information on the Web. XML has

become the new standard for semi-structured data exchange over the Web. In this paper, therefore, we study

the XML web warehouse and propose an approach to the problems of change detection and warehouse

maintenance in an XML web warehouse system. This paper has three major contributions. First, we propose

an object-oriented data model for XML web pages in the web warehouse as well as system architecture for

change detection and warehouse maintenance. Second, we propose a change detection method based on

mobile agent technology to actively detect changes of data sources of the web warehouse. Third, we propose

an incremental and deferred maintenance method to maintain XML web pages in the web warehouse. We

compared our approach with a rewriting approach to storage and maintenance of the XML web warehouse

by experiments. Performance evaluation shows that our approach is more efficient than the rewriting ap-

proach in terms of the response time and storage space of the web warehouse.

1 INTRODUCTION

The World Wide Web is a popular broadcast medium

that contains a huge amount of information. Infor-

mation on the Web is important not only to individ-

ual users but also to business organizations, espe-

cially for decision-making purposes. Therefore, how

to efficiently and effectively utilize web information

is an important issue. Recently, the concept of web

warehousing was proposed to address this issue (Ng,

1998; Xyleme, 2001). The idea of web warehousing

is to build a web warehouse, which materializes and

manages useful information from the Web, so as to

facilitate utilization of web information. A web

warehouse is a repository of web pages extracted

from remote web sites. It has specific data sources

and is built with a specific theme or purpose that is

of interest to a user community. Because XML has

become the new standard for semi-structured data

exchange over the Web, we study in this paper the

web warehouse of XML data, which will be called

the XML web warehouse thereafter.

If changes occur at data sources, a web ware-

use should be aware of those changes that may

affect the data in the web warehouse. How does a

web warehouse detect such changes is called the

change detection problem. In the traditional data

warehouse environment, data sources are cooperated

with the data warehouse and hence will send mes-

sages about their changes to the data warehouse

(Labio, 1995). However, data sources of a web

warehouse are autonomous web sites and will not

actively notify the web warehouse of their changes.

Previous work on change detection of web data que-

ries data sources for their changes (Bhowmick, 2000;

Lim, 2001), which requires more message transmis-

sion and is inefficient. Therefore, how to efficiently

detect changes of data sources is an important prob-

lem to a web warehouse.

If changes occur at data sources, a web ware-

use may need to be maintained. Deciding whether

and how to maintain a web warehouse is called the

warehouse maintenance problem. The problem of

warehouse maintenance is not a brand new problem

and has been studied for traditional data warehouses.

However, most of the previous work on this problem

was focused on the relational data model and the

traditional data warehouse environment (Agrawal,

1997; Zhuge, 1995; Zhuge, 1996). A web warehouse

can be differentiated from a traditional data ware-

house in terms of its data model and operating envi-

ronment, and hence requires a different maintenance

method. Therefore, how to efficiently maintain a

web warehouse is also an important problem to

Chao C. (2005).

CHANGE DETECTION AND MAINTENANCE OF AN XML WEB WAREHOUSE.

In Proceedings of the Seventh International Conference on Enterprise Information Systems, pages 52-59

DOI: 10.5220/0002555700520059

 SciTePress

study.

In this paper, we propose an approach to the

problems of change detection and warehouse main-

tenance for an XML web warehouse system. First,

we propose an object-oriented data model for XML

web pages in the web warehouse as well as system

architecture for change detection and warehouse

maintenance. Then, we propose a change detection

method based on mobile agent technology to ac-

tively detect changes of data sources of the web

warehouse. Finally, we propose an incremental and

deferred maintenance method to maintain XML web

pages in the web warehouse. We have implemented

an experimental prototype system for change detec-

tion and maintenance of an XML web warehouse

and have compared our approach with a rewriting

approach by experiments. Performance evaluation

shows that our approach is efficient in terms of the

response time and storage space of the web ware-

house.

The remainder of this paper is organized as fol-

lows. In Section 2 we illustrate the data model of the

web warehouse and the system architecture for

change detection and warehouse maintenance. In

Section 3 we present the change detection method

and algorithm. In Section 4 we present the ware-

house maintenance method and algorithm. Section 5

illustrates the experimental results and states our

observations from the experimental results. Section

6 concludes this paper and gives some directions for

future research.

2 DATA MODEL AND SYSTEM

ARCHITECTURE

2.1 Data Model

Our web warehouse stores XML data from the Web.

Hence, we propose a data model, called the XML

Web Warehouse Data Model (XWWDM), for XML

web pages in the web warehouse. Due to the hierar-

chical structure of an XML web page, we follow the

Document Object Model (Apparao, 1998) to de-

compose an XML web page into a tree structure.

Besides, the design of the XWWDM model is based

on the OEM-like model (Chawathe, 1999) and con-

siders the characteristics of a web warehouse. First,

a web warehouse is like a data warehouse in that it

can store historical data. Therefore, the data model

includes version information to keep track of the

change of data. Second, data in a web warehouse are

sourced from remote web sites. Therefore, the data

model includes source information to identify the

source of data. The XWWDM model is an ob-

ject-oriented model whose class definition is shown

in Figure 1.

class XML_Page {root: XML_Node, version:

Version_Info, source: Source_Info};

class XML_Node

{content: Node_Content, version: Version_Info};

class Node_Content {label: string, value: string, p-node:

XML_Node, child#: integer, s-action: char};

class Version_Info

{version#: integer, update-time: time};

class Source_Info

{url: string, title: string};

class Update

{content: Update_Content, source: Source_Info};

class Update_Content {label: string, value: string,

p-node: XML_Node, detect-time: time, action: char};

Figure 1: The class definition of the XWWDM model

An XML web page is represented as an object of

the class XML_Page, which has three attributes root,

version, and source. The attribute root records the

root node of the tree structure of the web page. The

attributes version and source record the newest ver-

sion information and source information of the web

page, respectively. Each node of a web page is rep-

resented as an object of the class XML_Node, which

has two attributes content and version. The attribute

content records the content, position, and source

action of a node. The attribute version records the

version information of a node. The class

Node_Content has five attributes label, value,

p-node, child#, and s-action. The attributes label and

value record the tag label and data content of a node,

respectively. The attributes p-node and child# record

the parent node and child number under its parent,

respectively. The attribute s-action records the

source action causing the creation of a node, whose

value is I (for insertion), D (for deletion), or M (for

modification). The class Version_Info has two attrib-

utes version# and update-time, which record the

version number and time of last update, respectively.

The class Source_Info has two attributes url and title,

which record the URL and title of the source web

page, respectively.

We adopt a change-centric approach to storage

of all versions of an XML web page. Only the first

version is completely stored. For subsequent ver-

sions, only deltas are stored. As shown in Figure 2,

all frames represent the same web page, in which

each frame represents a specific version at time Ti.

The first frame represents the first version, in which

all nodes of a web page are stored. Other frames

represent subsequent versions, in which only nodes

that are changed are stored. The number and letter

drawn by a node are the child number and source

CHANGE DETECTION AND MAINTENANCE OF AN XML WEB WAREHOUSE

action of the node. Different versions of a node have

the same parent node and child number but different

version numbers. The first version of a node has a

value I in the attribute s-action. A value D in the last

version of a node indicates that this node has been

deleted.

Figure 2: All versions of an XML web page

A source update is represented as an object of the

class Update, which has two attributes content and

source. The attribute content records a variety of

information about the update. The attribute source

records the identification information of the web

page in which the update occurs. The class Up-

date_Content has five attributes label, value, p-node,

detect-time, and action. The attributes label, value,

and p-node have similar meaning as in the class

Node_Content. The attributes detect-time and action

record the detection time and type of the update,

respectively. The update type can be I (for insertion),

D (for deletion) or M (for modification). Source up-

dates are detected by mobile agents and transmitted

from data sources to the web warehouse, and are

used for the purpose of maintaining the web ware-

house.

2.2 System Architecture

The system architecture for change detection and

maintenance of the XML web warehouse is shown

in Figure 3, which is divided into three layers: the

storage layer, the system kernel layer, and the mobile

agent layer. The primary components and their func-

tions in each of the three layers are presented below.

2.2.1 Storage Layer

This layer is where the web warehouse is stored. The

web warehouse uses the XWWDM model to store

specific and historical XML data from the Web. The

web warehouse serves as a knowledge base of deci-

sion support systems, providing data for web mining

and on-line analytical processing.

Mobi l e A gent Moni t or System

web warehouse

I nt egr at or

Update Storage

Management

Module

Message Module

Mobi l e Agent

Messag e

Module

De t e c t tio n

Module

Mobi l e A gent L ay er

System Kernel Layer

Storage Layer

Regi st er

Figure 3: System architecture

2.2.2 System Kernel Layer

This layer is the kernel of the part of the web ware-

house system for change detection and warehouse

maintenance. It includes the Mobile Agent Monitor

System, the Integrator, and the Update Storage. The

Mobile Agent Monitor System further consists of

two modules: the Management Module and the

Message Module. The Management Module is re-

sponsible for producing, dispatching, and tracing

mobile agents. The Message Module is responsible

for sending messages to and receiving messages

from mobile agents as well as placing received

source updates into the Update Storage. The Inte-

grator is responsible for maintaining the web ware-

house according to source updates stored in the Up-

date Storage. The Update Storage is a temporary

storage of source updates sent by mobile agents and

provides the Integrator with required update infor-

mation for warehouse maintenance.

2.2.3 Mobile Agent Layer

This layer includes mobile agents that are operating

in data sources of the web warehouse and are re-

sponsible for detecting and reporting changes of data

sources. Each mobile agent includes three compo-

nents: the Detection Module, the Message Module,

and the Register. The Detection Module is responsi-

ble for detecting changes of the data source. The

Message Module is responsible for sending mes-

sages to and receiving messages from the Mobile

Agent Monitor System. The Register is the storage

for the mobile agent to store the last states of web

pages and detected source updates.

time

ICEIS 2005 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

3 CHANGE DETECTION

In this section, we address the issue of change detec-

tion of data sources of an XML web warehouse. Be-

cause a web warehouse system is operating in the

Internet environment and its data sources are remote

and autonomous web sites, it is better for it to ac-

tively detect changes of data sources so as to speed

up data refreshment and reduce network traffic. Mo-

bile agents have the feature of cross platform as well

as the abilities of active detection and rapid report-

ing. Therefore, we propose a change detection

method based on mobile agent technology in order

for the web warehouse system to fully and effi-

ciently obtain changes of data sources. The proce-

dure of our change detection method consists of the

following four steps:

1. The Mobile Agent Monitor System produces

and dispatches a mobile agent to each of the

data sources.

2. When a mobile agent arrives at a data source, it

first sends source information of the data source

back to the Mobile Agent Monitor System. This

source information allows the Mobile Agent

Monitor System to trace the mobile agent.

3. Thereafter a mobile agent will actively and pe-

riodically detect changes of a data source using

the Change Detection Algorithm. If changes are

detected, the mobile agent sends detected

source updates to the Mobile Agent Monitor

System.

4. When the Mobile Agent Monitor System re-

ceives source updates from a data source, it

stores these updates in the Update Storage for

the purpose of warehouse maintenance.

The Detection Module of a mobile agent uses the

Change Detection Algorithm, the CD algorithm for

short, shown in Figure 4 to detect the difference be-

tween the current state and the last detected state of

an XML web page and send a collection of updates

to the Mobile Agent Monitor System through the

Message Module. Here we first give a brief over-

view of the CD algorithm. First, the algorithm

fetches an XML web page from the web site and

finds the last recorded state of the web page from the

posed into tree structures using an XML parser. Then,

these two tree structures are compared level-by-level

in a top-down fashion to detect their difference.

During this process, a collection of updates repre-

senting the difference is accumulated. Finally, the

current state of the web page is recorded in the Reg-

ister and the collection of updates is sent to the Mo-

bile Agent Monitor System. Now we explain the CD

algorithm in detail. The algorithm executes the fol-

lowing steps in sequence:

Algorithm Change Detection Algorithm

¾ Fetch an XML web page WP.

¾ Find the last recorded state LP of WP from the

¾ Initialize a collection of updates UC to be empty.

¾ Use an XML parser to decompose WP and LP

into tree structures.

¾ For each level L of WP and LP do

c For each node WPN of level L of WP do

Find the node LPN of level L of LP such that

LPN.content.label = WPN.content.label

If LPN is found then Mark WPN and LPN.

If WPN.content.value ≠ LPN.content.value then

⇒ Create an update object U whose attributes are

as follows:

content.label ← the tag label of WPN

content.value ← the data content of WPN

content.p-node ← the parent node of WPN

content.detect-time ← current system time

content.action ← ‘M’

source.url ← the URL of WP

source.title ← the title of WP

⇒ Put U into UC.

d For each node WPN of level L of WP do

If WPN is not marked then

⇒ Create an update object U whose attributes are

as follows:

content.label ← the tag label of WPN

content.value ← the content of WPN

content.p-node ←

the parent node of WPN

content.detect-time ← current system time

content.action ← ‘I’

source.url ← the URL of WP

source.title ← the title of WP

⇒ Put U into UC.

e For each node LPN of level L of LP do

If LPN is not marked then

⇒ Create an update object U whose attributes are

as follows:

content.label ← the tag label of LPN

content.value ← the data content of LPN

content.p-node ← the parent node of LPN

content.detect-time ← current system time

content.action ← ‘D’

source.url ← the URL of LP

source.title ← the title of LP

⇒ Put U into UC.

¾ Record the state of WP in the Register.

¾ Send UC to the Mobile Agent Monitor System.

End Algorithm

Figure 4: The change detection algorithm

1. Fetch an XML web page WP from the web site.

WP is the current state of the web page whose

difference from its last recorded state is to be

detected.

2. Find the last recorded state LP of WP from the

CHANGE DETECTION AND MAINTENANCE OF AN XML WEB WAREHOUSE

sumed to have been stored in the web ware-

house, it must have the last recorded state.

3. Initialize a collection of updates UC to be

empty. UC represents the difference between

WP and LP.

4. Decompose WP and LP into tree structures us-

ing an XML parser.

5. Compare each level of the tree structures of WP

and LP in a top-down fashion in a loop. For

each level L of WP and LP, there are three inner

loops. In the first inner loop, for each node

WPN of level L of WP, find the node LPN of

level L of LP such that the tag label of LPN is

equal to that of WPN. If such a node is found,

mark WPN and LPN to indicate that this node

exists in both states of the web page. If the data

content of WPN is not equal to that of LPN,

which indicates that WPN has been modified,

create a modification update and put it into UC.

In the second inner loop, for each node WPN of

level L of WP, check if it is marked. If it is not

marked, which indicates that this node has been

inserted to the current state, create an insertion

update and put it into UC. In the third inner

loop, for each node LPN of level L of LP, check

if it is marked. If it is not marked, which indi-

cates that this node has been deleted from the

last state, create a deletion update and put it into

UC.

6. Record the state of WP in the Register. This

state will become the last recorded state next

time this web page is detected.

7. Send the accumulated collection of updates UC

to the Mobile Agent Monitor System.

4 WAREHOUSE MAINTENANCE

In this section, we address the issue of maintaining

an XML web warehouse in response to changes of

data sources. The way to warehouse maintenance

somewhat depends on the way to storage of the

warehouse as well as the way to change detection of

data sources. Our web warehouse adopts a

change-centric approach to storage of historical data

of web pages. Our change detection method utilizes

mobile agents to send actively and periodically

source updates back to the web warehouse. Accord-

ingly, we propose an incremental and deferred

maintenance method to maintain XML web pages in

the web warehouse. Incremental maintenance of a

web page means that only parts of the web page are

updated according to its source updates. It is gener-

ally more efficient in terms of time and space than

rewriting the whole web page. Deferred maintenance

of a web page means that the web page is main-

tained only when it is accessed. It is generally more

time efficient than immediate maintenance in which

a web page is maintained immediately after a source

update occurs. Besides, the user is still able to access

reasonably up-to-date web pages in a relatively short

time. The procedure of our warehouse maintenance

method consists of the following four steps:

1. When the web warehouse receives a request

from a user to access a web page, it notifies the

Integrator to maintain the web page.

2. The Integrator checks the Update Storage to see

if there are source updates relevant to the web

page. The relevant source updates are those that

have the same source location information (i.e.,

URL) as the web page. All updates made to the

source web page of this web page are consid-

ered as relevant source updates.

3. If the web page has relevant source updates,

which indicates that it is out of date, the Inte-

grator maintains it using the Web Warehouse

Maintenance Algorithm.

4. The web warehouse returns the requested web

page to the user.

The Integrator uses the Web Warehouse Mainte-

nance Algorithm, the WWM algorithm for short,

shown in Figure 5 to maintain an XML web page in

the web warehouse according to the relevant source

updates of the web page in the Update Storage. Here

we first give a brief overview of the WWM algo-

rithm. First, the algorithm finds the XML web page

to be maintained in the web warehouse. Then, it gets

and removes the relevant source updates of the web

page from the Update Storage. These relevant up-

dates are sorted by their detection time to reflect

their order of occurrence. Finally, it maintains the

web page for each relevant update in sequence. A

new version of the web page is obtained. Now we

explain the WWM algorithm in detail. The algorithm

executes the following steps in sequence:

1. Find the XML web page to be maintained WP

in the web warehouse.

2. Get and remove a collection of relevant source

updates RU from the Update Storage. RU is the

collection of relevant source updates of WP.

3. Sort the updates in RU by their detection time

recorded in the attribute detect-time to reflect

their order of occurrence.

4. Create a version object V whose attribute ver-

sion# is the current version number of WP plus

one and whose attribute update-time is the cur-

rent system time. V records the newest version

information and will be stored in the next ver-

sion of WP.

5. Change the newest version information of WP

to V.

ICEIS 2005 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

Algorithm Web Warehouse Maintenance Algorithm

¾ Find the XML web page to be maintained WP

in the web warehouse.

¾ Get and remove a collection of relevant source

updates RU from the Update Storage.

¾ Sort the updates in RU by the attribute detect-time.

¾ Create a version object V whose attributes are

as follows:

version# ← WP.version.version# + 1

update-time ← current system time

¾ WP.version ← V

¾ For each update U of RU do

c If U.content.action = ‘I’ then

⇒ Find the maximum child# MN among all nodes

whose parent is U.content.p-node

⇒ Insert a node whose attributes are as follows:

content.label ← U.content.label

content.value ← U.content.value

content.p-node ← U.content.p-node

content.child# ← MN + 1

content.s-action ← ‘I’

version ← V

d If U.content.action = ‘D’ then

⇒ Find the newest version of node N such that

.content.p-node = U.content.p-node and

N.content.label = U.content.label

⇒ Insert a node whose attributes are as follows:

content.label ← N.content.label

content.value ← N.content.value

content.p-node ← N.content.p-node

content.child# ← N.content.child#

content.s-action ← ‘D’

version ← V

e If U.content.action = M then

⇒ Find the newest version of node N such that

N.content.p-node = U.content.p-node and

N.content.label = U.content.label

⇒ Insert a node whose attributes are as follows:

content.label ← N.content.label

content.value ← U.content.value

content.p-node ← N.content.p-node

content.child#

← N.content.child#

content.s-action ← ‘M’

version ← V

End Algorithm

Figure 5: The warehouse maintenance algorithm

6. Maintain WP for each update U of RU in se-

quence in a loop. For each update U, check to

see if U is an insertion, a deletion, or a modifi-

cation. If U is an insertion, insert a node whose

parent node is U.content.p-node to WP. The at-

tribute child# of the inserted node is the current

maximum child number among its siblings plus

one. This inserted node represents that a node

was inserted to the source web page of WP. If U

is a deletion, find the newest version of the

node in WP that corresponds to the node that

was deleted from the source web page of WP

and insert a node to WP. This inserted node

represents that a node was deleted from the

source web page of WP. If U is a modification,

find the newest version of the node in WP that

corresponds to the node that was modified in

the source web page of WP and insert a node to

WP. This inserted node represents that a node

was modified in the source web page of WP

and becomes the newest version of the node in

WP. After every update of RU is processed, a

new version of WP is obtained.

5 PERFORMANCE EVALUATION

We have implemented an experimental prototype

system for change detection and maintenance of an

XML web warehouse. The web warehouse is built

on the ObjectStore object-oriented database man-

agement system and programs are written in the Java

object-oriented programming language. Besides, we

adopt the Aglets Software Development Kit (ASDK)

and the Aglets Workbench both from IBM as the

development tool and operating environment of mo-

bile agents, respectively. The hardware platform

consists of several personal computers that commu-

nicate with the Internet.

In the performance evaluation, we compare our

approach with a rewriting approach to storage and

maintenance of an XML web warehouse. In the re-

writing approach, every version of a web page is

completely stored. In our approach, on the other

hand, every version except the first one stores only

its difference from the previous version. As in our

approach, the rewriting approach uses the deferred

strategy to maintain a web page. However, it adopts

a different method for change detection and ware-

house maintenance. While receiving an access re-

quest for a web page, the web warehouse sends a

request to the source web site of the web page for

the current state of the web page. After maintaining

the web page by creating a complete version of the

current state, the web warehouse returns the re-

quested web page to the user. In our approach, on

the other hand, the requested web page is maintained

using updates that have already been sent back by

the mobile agent and stored in the local Update

Storage. We compare our approach with the rewrit-

ing approach in terms of two important performance

criteria: the response time and the storage space.

The response time is the elapsed time starting

from the web warehouse receives an access request

until it returns the requested web pages. From the

point of view of the user of a web warehouse, the

CHANGE DETECTION AND MAINTENANCE OF AN XML WEB WAREHOUSE

response time is the most important criterion to

judge the performance of the system. We separately

consider two factors that may affect the response

time, the number of updates to a web page and the

number of accessed web pages. We first compare the

response time of two approaches for accessing a

single web page with different numbers of updates to

the web page. An experimental result of such a

comparison is illustrated in Figure 6(a). In Figure 6

and Figure 7, the abbreviations IM (standing for

incremental maintenance) and RW (standing for re-

writing) represent our approach and the rewriting

approach, respectively. From the experimental re-

sults of this comparison, we observe two phenomena.

First, the response time of our approach increases

with the number of updates to a web page. However,

the response time of the rewriting approach does not

depend on the number of updates. Second, the re-

sponse time of our approach is shorter than that of

the rewriting approach in general, especially when

the number of updates in smaller.

(a)

(b)

Figure 6: Comparison of the response time

We also compare the response time of two ap-

proaches for accessing multiple web pages. In this

comparison, the number of updates to each web page

is fixed and the numbers of updates to these web

pages are small. An experimental result of such a

comparison is illustrated in Figure 6(b). From the

experimental results of this comparison, we observe

two phenomena. First, the response time of both

approaches increases with the number of accessed

web pages. Second, no matter how many web pages

are accessed, the response time of our approach is

always shorter than that of the rewriting approach as

long as the numbers of updates to web pages are

small.

The storage space is the size of the secondary

storage required for storing the historical data of a

web page in the web warehouse. We consider two

factors that may affect the storage space, the size of

a web page and the update percentage of a web page.

We compare the storage space of two approaches in

terms of three different sizes (large, median, and

small) and three different update percentages (30%,

50%, and 80%). An experimental result of such a

comparison is illustrated in Figure 7, which shows

the ratio of the storage space of our approach to the

storage space of the rewriting approach. From the

experimental results of this comparison, we observe

three phenomena. First, the storage space of our ap-

proach is smaller than that of the rewriting approach

in most of the situations. Second, the storage space

of our approach increases with the update percentage

of the web page. Third, our approach is more advan-

tageous to web pages of larger size. It tends to re-

quire more storage space than the rewriting approach

as the size of the web page gets smaller and the up-

date percentage of the web page gets higher.

Accessing a single web page

100

200

300

400

500

600

700

2 5 10 20 40

Number of updates

Response time (

Stora

e Space

20%

40%

60%

80%

100%

120%

140%

large median small

e Si ze

Ratio of storage space

I M( 30%)

I M (50%)

I M( 80%)

Accessing multiple web pages

5000

10000

15000

2 5 10 15 20 30

Number of web pages

Response time(

Figure 7: Comparison of the storage space

6 CONCLUSION AND FUTURE

WORK

In this paper, we proposed an approach to change

ICEIS 2005 - DATABASES AND INFORMATION SYSTEMS INTEGRATION

detection and warehouse maintenance for an XML

web warehouse system. We compared our approach

with a rewriting approach in terms of the response

time and storage space of the web warehouse. In our

approach, mobile agents are dispatched by the web

warehouse to data sources and will actively and pe-

riodically detect and report changes of data sources

back to the web warehouse. Beside, the web ware-

house is incrementally maintained in response to

source updates that have already been stored in the

local storage. These can dramatically reduce the

number of messages transmitted between the web

warehouse and data sources. Therefore, our ap-

proach is more efficient than the rewriting approach

in terms of the response time. With regard to the

storage space, our approach adopts a change-centric

approach in which every version of a web page ex-

cept the first version stores only its difference from

the previous version. This can dramatically reduce

the size of the storage required for historical data of

web pages in the web warehouse. Therefore, our

approach is more efficient than the rewriting ap-

proach in terms of the storage space.

In the future, we will improve our approach in an

attempt to further increase its efficiency. First, with

regard to the storage of all versions of an XML web

page, we will consider storing the current version

completely and only deltas for historical data. Sec-

ond, we will try to improve our warehouse mainte-

nance algorithm so as to handle large number of

updates more efficiently. Besides, more comprehen-

sive experimentation will be performed to extract

conclusive results.

REFERENCES

Agrawal, D., El Abbadi, A., Singh, A., Yurek, T., 1997.

Efficient view maintenance at data warehouses. In

Proceedings of the 1997 ACM SIGMOD International

Conference on Management of Data, pp. 417-427.

Apparao, V., 1998. Document Object Model (DOM) Level

1 Specification (Version 1.0).

Bhowmick, S. S., Ng, W. K., Madria, S. K., Lim, E. P.,

2000. Detecting and representing relevant web deltas

using web join. In Proceedings of the 20

IEEE In-

ternational Conference on Distributed Computing Sys-

tems, pp. 255-262.

Chawathe, S. S., Abiteboul, S., Widom, J., 1999. Manag-

ing historical semistructured data. Theory and Practice

of Object Systems, Vol. 5, No. 3, pp. 143-162.

Labio, W., Garcia-Molina, H., 1995. Efficient snapshot

differential algorithm for data warehousing. In Pro-

ceedings of the 22

International Conference on Very

Large Data Bases, pp. 63-74.

Lim, S. J., Ng, Y. K., 2001. An automated change detec-

tion algorithm for HTML documents based on seman-

tic hierarchies. In Proceedings of the 17

IEEE Inter-

national Conference on Data Engineering, pp.

303-312.

Ng, W. K., Lin, E. P., Huang, C. T., Bhowmick, S., Qin, F.

Q., 1998. Web warehousing: an algebra for web in-

formation. In Proceedings of the 1998 IEEE Forum on

Research and Technology Advances in Digital Librar-

ies, pp. 228-237.

Xyleme, L., 2001. A dynamic warehouse for XML data of

the web. IEEE Data Engineering Bulletin, Vol. 24, No.

2, pp. 40-47.

Zhuge, Y., Garcia-Molina, H., Hammer, J., Widom, J.,

1995. View maintenance in a warehousing environ-

ment. In Proceedings of the 1995 ACM SIGMOD In-

ternational Conference on Management of Data, pp.

316-327.

Zhuge, Y., Garcia-Molina, H., Wiener, J. L., 1996. The

Strobe algorithms for multi-source warehouse consis-

tency. In Proceedings of the 4

IEEE International

Conference on Parallel and Distributed Information

Systems, 146-157.

CHANGE DETECTION AND MAINTENANCE OF AN XML WEB WAREHOUSE