0167-9236/$ - see front matter D 2005 Elsevie

doi:10.1016/j.dss.2005.05.012

* Corresponding author. Tel.: +852 2609 82

5505.

E-mail address: yang@se.cuhk.edu.hk (C.C

www.elsevier.com/locate/dss

cision Support System

An information delivery system with automatic summarization

for mobile commerce

Christopher C. Yang a,*, Fu Lee Wang b

aDepartment of Systems Engineering and Engineering Management, The Chinese University of Hong Kong, Shatin, Hong Kong, ChinabDepartment of Computer Science, City University of Hong Kong, Kowloon Tong, Hong Kong, China

Available online 14 July 2005

Abstract

Wireless access with handheld devices is a promising addition to the WWW and traditional electronic business. Handheld

devices provide convenience and portable access to the huge information space on the Internet without requiring users to be

stationary with network connection. Many customer-centered m-services applications have been developed. The mobile

computing, however, should be extended to decision support in an organization. There is a desire of accessing most update

and accurate information on handheld devices for fast decision making in an organization. Unfortunately, loading and

visualizing large documents on handheld devices are impossible due to their shortcomings. In this paper, we introduce the

fractal summarization model for document summarization on handheld devices. Fractal summarization is developed based on

the fractal theory. It generates a brief skeleton of summary at the first stage, and the details of the summary on different levels of

the document are generated on demands of users. Such interactive summarization reduces the computation load in comparing

with the generation of the entire summary in one batch by the traditional automatic summarization, which is ideal for wireless

access. The three-tier architecture with the middle-tier conducting the major computation is also discussed. Visualization of

summary on handheld devices is also investigated. The automatic summarization, the three-tier architecture, and the information

visualization are potential solutions to the existing problems in information delivery to handheld devices for mobile commerce.

D 2005 Elsevier B.V. All rights reserved.

Keywords: Document summarization; Mobile commerce; Fractal summarization; Handheld devices; Financial news delivery

1. Introduction

The advance of mobile network creates business

opportunities and provides value-added services to

users. Access to the Internet through mobile phones

r B.V. All rights reserved.

39; fax: +852 2603

. Yang).

and other handheld devices is growing significantly in

recent years. Many m-services applications have been

developed for the handheld devices [5–8,38]. How-

ever, most current applications are customer-centered

applications, for example, users can now surf the web,

check e-mail, read news, and quote stock price, etc.,

using handheld devices. The mobile computing

should not be limited to user-centered applications

only. In this age of information, mobile applications

De s 43 (2007) 46–61

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 47

should be extended to decision support in an organi-

zation. With a fast paced economy, organizations need

to make decisions as fast as possible, they must gain

competitive advantage by having access to the most

current and accurate information available. For

instance, a huge amount of financial news is generated

everyday, and access to update financial information

is important during decision making. On the other

hand, the executives of an organization need make

decision when they are on the road. As a result, there

is an urgent need of information access through hand-

held devices.

There are many shortcomings associated with

handheld devices although the development of hand-

held devices is fast in the recent years. The short-

comings include limited screen size with low

resolution, low bandwidth, and low memory capacity.

It is impossible to search and visualize the critical

information on a small screen with an intolerable slow

downloading speed using handheld devices. Auto-

matic summarization summarizes a document for

users to preview its major content. Users may deter-

mine if the information fits their needs by reading the

summary instead of browsing the whole document

one by one. The amount of information displayed

and downloading time are significantly reduced.

Active researches on automatic summarization have

been carried out. Summarization techniques have been

applied to delivery of information on handheld

devices [3,42]. Traditional summarizations do not

consider the hierarchical structure of document but

consider the document as a sequence of sentences.

Most traditional summarization systems extract sen-

tences from the source document and concatenate

together as summary. However, it is believed that

the document summarization on handheld devices

must make use of btree viewQ [7] or bhierarchicaldisplayQ [32]. Similar techniques have been applied

to web browsing, an outline processor organizes the

web page in a tree structure and the user click the link

to expand the subsection and view the detail [4].

Hierarchical display is suitable for navigation of a

large document and it is ideal for small area display.

Therefore, a new summarization model with hierarch-

ical display is required for summarization on handheld

devices. Summarization of web pages on handheld

devices has been investigated [5–8]. However, a

large document exhibits totally different characteris-

tics from web pages. A web page usually contains a

small number of sentences that are organized into

paragraphs, but a large document contains much

more sentences that are organized into a more com-

plex hierarchical structure [48,49]. Besides, the sum-

marization on webpage is mainly based on thematic

features only [6]. However, it has been proved that

other document features play a role as important as the

thematic feature [10,22]. Therefore, a more advance

summarization model combined with other document

features is required for browsing of large document

and other information sources on handheld devices

[49–51]. With powerful summarization tool, the abil-

ity of handheld devices will be greatly enhanced.

Visualization of various information sources becomes

feasible, for example, financial news delivery to hand-

held devices. It provides an information visualization

tool for m-commerce.

The information access through mobile devices has

drawn attention from researchers in recent years.

Mobile devices support instant information and data

access that is capable to support decision making. For

example, Mendelsohn [35] has presented how physi-

cians may request information to answer questions

that occur at the point-of-care through mobile devices.

A number of researchers have worked on advanced

text input entry for handheld device [29,30,39]. More

specifically, there are researches that focus on naviga-

tion of documents on mobile devices. Lamming et al.

[26] have presented the Satchel system that used

tokens to represent documents on mobile device and

developed a context-sensitive user interface. In this

work, we focus on the delivery of information to

handheld devices with the support of automatic sum-

marization. Example of how the proposed system

supports the delivery of Yahoo! financial information

is presented.

The paper is organized as follows. Section 2 dis-

cusses the shortcomings of handheld devices and

information visualization on handheld devices. The

three-tier architecture, which reduces the computing

load of the handheld devices, is used. Section 3

proposes the fractal summarization model based on

the statistical data and the hierarchical structure of

documents. Thematic, location, heading and cue fea-

tures are adopted. Experiments have been conducted

and the results show that the fractal summarization

outperforms the traditional summarization. Fractal

Web Server

HTML

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6148

summarization is further extended to information

browsing on handheld devices. Section 4 will demon-

strate the financial news delivery on handheld devices.

DB Server forSummarizer User’s PC

Sum

mar

izer

Bro

wse

r

User

SQL

Document Server

XML

Fig. 2. Document browsing with summarizer on PC.

2. Document delivery architecture on handheld

devices

Traditionally, two-tier architecture is typically uti-

lized for Internet access. The user’s PC connects to the

Internet directly, and the content loaded will be fed to

the web browser and present to the user as illustrated

in Fig. 1.

However the information available online has

increased explosively, it results in a well-recognized

problem of information overloading. Advance-search-

ing techniques solve the problem by filtering most of

the irrelevant information. However, the precision of

most of the commercial search engines is not high.

Users may find only a few relevant documents out of

a large pool of searching result. Due to the huge

volume of documents, it will take a lot of time for

the users to browse the searching result one by one

and identify the relevant information using desktop

computers. Automatic summarization summarizes a

document for users to preview its major content.

Users may determine if the information fits their

needs by reading their summary instead of browsing

the whole document one by one. The amount of

information displayed and downloading time are sig-

nificantly reduced. An automatic summarizer is there-

fore introduced to summarize a document for the users

to preview before presenting the whole document. As

shown in Fig. 2, the content will be first fed to the

summarizer after loading to the user’s PC. The sum-

marizer connects to database server when necessary

and generates a summary to display on the browser.

The convenience of handheld devices allows infor-

mation access without geometric limitation; however,

Web Server

DB Server for Summarizer WAP Gateway

HTML

SQ

Document Server

XML

Local Synchronization

User

Wireless

Handhel

WML

PDA

Sum

mar

izer

B

row

ser

Fig. 3. Document browsing with summarizer on WAP.

User Web Server

User’s PC

Bro

wse

r

Fig. 1. Document browsing on PC.

there are other limitations of handheld devices that

restrict their capability. Although the development of

wireless handheld devices is fast in recent years, there

are many shortcomings associated with these devices,

such as screen size, bandwidth, and memory capacity.

There are two major categories of wireless handheld

devices, namely WAP-enabled mobile phones and

wireless PDAs. At present, the typical display size

of popular WAP-enabled handsets and PDAs are rela-

tively small in comparison with a standard PC. The

memory capacity of a handheld device greatly limits

the amount of information that can be stored. A large

document cannot be entirely downloaded to the hand-

held device and present to user directly. The current

bandwidth available for WAP is relatively narrow. It is

not comparable with the broadband internet connec-

tion for PC. The handheld devices impose other con-

straints that do not exist on desktop computers. The

two-tier architecture cannot be applied on handheld

devices since the computing power of handheld

devices is insufficient to perform summarization and

the network connection of mobile network does not

provide sufficient bandwidth for navigation between

the summarizer and other servers.

The three-tier architecture as illustrated in Fig. 3 is

proposed. A WAP gateway is setup to conduct the

summarization. The WAP gateway connects to Inter-

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 49

net trough broadband network. The wireless handheld

devices can conduct interactive navigation with the

gateway through wireless network to retrieve the

summary piece by piece. Alternatively, if the PDA

is equipped with more memory, the complete sum-

mary can be downloaded to PDA through local

synchronization.

3. Automatic summarization

3.1. Traditional summarization

Traditional automatic text summarization is the

selection of sentences from the source document

based on their significance to the document [10,28].

The selection of sentences is conducted based on the

salient features of the document. The thematic, loca-

tion, heading, and cue features are the most widely

used summarization features.

! The thematic feature is first identified by Luhn

[28]. Edmundson proposed to assign the thematic

weight to keyword based on term frequency, and

the sentence thematic score as the sum of thematic

weight of constituent keywords [10]. In informa-

tion retrieval, absolute term frequency by itself is

considered as less useful than term frequency nor-

malized to the document length and term frequency

in the collection [18]. As a result, the tfidf (Term

Frequency, Inverse Document Frequency) method

is proposed to calculate the thematic weight of

keyword [41].

! The significance of sentence is indicated by its

location [2] based on the hypotheses that topic

sentences tend to occur at the beginning or in the

end of documents or paragraphs [10]. Edmundson

proposed to assign positive weights to sentences

as sentence location score according to their

ordinal position in the document, i.e., the sen-

tences in the first and last paragraphs and the first

and last sentences of the paragraphs. There are

several functions proposed to calculate the loca-

tion weight of sentences. Alternatively, the pre-

ference of sentence location can be stored in a

list called Optimum Position Policy, and the sen-

tences will be selected base on their order in the

list [27].

! The heading feature is proposed based on the

hypothesis that the author conceives the heading

as circumscribing the subject matter of the docu-

ment. When the author partitions the document into

major sections, he summarizes them by choosing

appropriate headings [10]. The formulation of

heading weight is very similar to the thematic

feature. A heading glossary is a list consisting of

all the words in headings and subheadings. Positive

weights are assigned to the heading glossary, where

the heading words will be assigned a weight rela-

tively prime to the subheading words. The sentence

heading score of sentence is calculated by the sum

of heading weight of its constituent words.

! The cue phrase feature is proposed by Edmundson

[10] based on the hypothesis that the probable

relevance of a sentence is affected by the presence

of pragmatic words such as bsignificantQ,bimpossibleQ, and bhardlyQ. A pre-stored cue dic-

tionary is used to identify the cue phases, which

comprises of three sub-dictionaries: (i) bonus

words, that are positively relevant; (ii) stigma

words, that are negatively relevant; and (iii) null

words, that are irrelevant. The sentence cue score

of sentence is calculated by the sum of cue weight

of its constituent words.

Typical summarization systems select a combina-

tion of summarization features [10,25,27], the total

sentence significance score (SSS) is calculated as,

SSS ¼ a1 � SSthematic þ a2 � SSlocation þ a3

� SSheading þ a4 � SScue ð1Þ

where SSthematic, SSlocation, SSheading and SScue are

sentence scores based on thematic feature, location

feature, heading feature and cue phrase feature,

respectively; and a1, a2, a3, and a4 are positive inte-

gers to adjust the weighting of four summarization

features. The sentences with sentence significant score

higher than a threshold are selected as part of the

summary. It has been proved that the weighting of

different summarization features does not have any

substantial effect on the average precision [25]. In our

experiment, the maximum score of each feature is

normalized to one, and the sentence significant score

is calculated as the sum of scores of all summarization

features without weighting.

Fig. 4. Koch curve at different abstractions level.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6150

3.2. Fractal theory and fractal view for controlling

information displayed

Fractals are mathematical objects that have high

degree of redundancy [31]. These objects are made

of transformed copies of themselves or part of them-

selves (Fig. 4). Mandelbrot was the first researcher

who investigated the fractal geometry and developed

the fractal theory [31]. In his well-known example,

the length of the British coastline depends on mea-

surement scale. The larger the scale is, the smaller

value of the length of the coastline is and the higher

the abstraction level is. The British coastline

includes bays and peninsulas. Bays include sub-

bays and peninsulas include sub-peninsulas. Using

fractals to represent these structures, abstraction of

the British coastline can be generated with different

abstraction degrees. Fractal theory is grounded in

geometry and dimension theory. Fractals are inde-

pendent of scale and appear equally detailed at any

level of magnification. Such property is known as

self-similarity. Any portion of a self-similar fractal

curve appears identical to the whole curve. If we

shrink or enlarge a fractal pattern, its appearance

remains unchanged.

Fig. 5. Fractal view for logical tree

Fractal view is a fractal-based method for control-

ling information displayed [23]. Fractal view provides

an approximation mechanism for the observer to

adjust the abstraction level and therefore control the

amount of information displayed. At a lower abstrac-

tion level, more details of the fractal object can be

viewed.

A physical tree is one of the classical examples of

fractal objects. A tree is made of a lot of sub-trees;

each of them is also a tree. By changing the scale, the

different levels of abstraction views are obtained (Fig.

5). The idea of fractal tree can be extended to any

logical tree. The degree of importance of each node is

represented by its fractal value. The fractal value of

root is 1, it propagated to other nodes with the follow-

ing expression:

Fvroot ¼ 1

Fvchild mode of x ¼ C Fvx

N1=Dx

(ð2Þ

where Fvx is the fractal value of node x; C is a

constant between 0 and 1 to control rate of decade;

Nx is the number of child nodes of node x; and D is

the fractal dimension. Mandelbrot had shown that the

fractal dimension of a tree is 1, because its total length

is finite and positive [31]. To simplify our discussion,

both C and D are considered to be 1.

A threshold value is chosen to control the amount

of information displayed, the nodes with a fractal

value less than the threshold value will be hidden

(Fig. 6). By changing the threshold value, the user

can adjust the amount of information displayed.

3.3. Fractal summarization

Advance summarization techniques take the docu-

ment structure into consideration to compute the prob-

ability of a sentence to be included in the summary.

Many studies of human abstraction process had

shown that the human abstractors extract the topic

at different abstraction level.

Fig. 6. An example of the propagation of fractal values.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 51

sentences according to the document structure from

the top level to the low level until sufficient informa-

tion has been extracted [11,15]. However, most tradi-

tional automatic summarization models consider the

source document as a sequence of sentences but

ignoring the structure of document. Some summariza-

tion systems may calculate sentence weight partially

based on the document structure, but they extract

sentences in a linear space. None of the current sum-

marization model is developed on the foundation of

the hierarchical structure of documents. Fractal Sum-

marization Model is proposed here to generate sum-

mary based on document structure. Fractal

summarization generates a brief skeleton of summary

at the first stage, and the details of the summary on

different levels of the document are generated on

demands of users. Such interactive summarization

reduces the computation load in comparing with the

Subsection

Paragraph Paragraph

Sentence Sentence

Term Term ...

Word Word ...

Character Character ...

Fig. 7. Prefractal struct

generation of the entire summary in one batch by the

traditional automatic summarization, which is ideal

for m-commerce.

Fractal summarization is developed based on the

fractal theory. In fractal summarization, the important

information is captured from the source text by explor-

ing the hierarchical structure and salient features of the

document. A condensed version of the document that is

informatively close to the original document is pro-

duced iteratively using the contractive transformation

in the fractal theory. Similar to the fractal geometry

applying on the British coastline where the coastline

includes bays, peninsulas, sub-bays, and sub-peninsu-

las, large document has a hierarchical structure with

several levels, chapters, sections, subsections, para-

graphs, sentences, and terms. A document is consid-

ered as prefractal that are fractal structures in their

early stage with finite recursion only [12]. A document

can be represented by a hierarchical structure as shown

in Fig. 7. A document consists of chapters. A chapter

consists of sections. A section may consist of subsec-

tions. A section or subsection consists of paragraphs. A

paragraph consists of sentences. A sentence consists of

terms. A term consists of words. A word consists of

characters. A document structure can be considered as a

fractal structure. At the lower abstraction level of a

document, more specific information can be obtained.

Although a document is not a true mathematical fractal

object since a document cannot be viewed in an infinite

abstraction level, we may consider a document as a

prefractal. The smallest unit in a document is character;

Document

Chapter Chapter ...

Section Section

Subsection ...

...

...

ure of document.

6. Set the current node to the root of thefractal tree.

7. Repeat7.1 Foreach child node under current node,

Calculate the fractal value of childnode.

7.2 Allocate Quota to child nodes in pro-portion to fractal values.

7.3 For each child nodes,If thequota is less than threshold value

Select the sentences in therange-block by extraction

Else

SetthecurrentnodetothechildnodeRepeat Step 7.1, 7.2, 7.3

8. Until all the child nodes under currentnode are processed

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6152

however, neither a character nor a word will convey

any meaningful information concerning the overall

content of a document. The lowest abstraction level

in our consideration is a term.

The Fractal Summarization Model applies techni-

ques from fractal view and fractal image compres-

sion [1,21]. In fractal image compression, an image

is regularly segmented into sets of non-overlapping

square blocks, called range-blocks, and then each

range-block is sub-divided into subrange-blocks,

until a contractive mapping can be found to represent

this subrange-block. The Fractal Summarization

Model generates the summary by a recursive deter-

ministic algorithm based on the iterated representa-

tion of a document. The original document is

represented as fractal tree structure according to its

hierarchical document structure. Then, the system

calculates the fractal value of each node and allo-

cates the sentence quota base on the fractal value.

The calculation of fractal value will be discussed

later in this section.

Given a document, a user can specify compression

ratio to indicate the amount of information displayed.

The summarization system calculates the number of

sentences to be extracted as summary accordingly and

the system assigns the number of sentences to the root

of document tree as the quota of sentences. The quota

of sentences is allocated to child nodes by propaga-

tion, i.e., the quota of parent node is shared by its

child nodes directly proportional to the fractal value of

the child nodes. The quota is then iteratively allocated

to child nodes of child nodes until the quota allocated

is less than a threshold value and the range-block can

be transformed to some key sentences by traditional

summarization methods (Fig. 8). The detail of the

Fractal Summarization Model is shown as the follow-

ing algorithm:

Fractal Summarization Model

1. Choose a Compression Ratio.2. Choose a Threshold Value.3. Calculate the Sentence Number Quota

of the summary.4. Divide the document into range-blocks.5. Transform the document into fractal

tree.

The compression ratio of summarization is defined

as the ratio of number of sentences in the summary to

the number of sentences in the source document. It

was chosen as 25% in most literatures because it has

been proved that extraction of 20% sentences can be

as informative as the full text of the source document

[36], those summarization systems can achieve up to a

96% precision [10,22,44]. However, Teufel pointed

out the high-compression ratio abstracting is more

useful, and 49.6% of precision is reported at 4%

compression ratio [43,44]. In order to minimize the

bandwidth requirement and reduce the pressure on

computing power of handheld devices, the default

value of compression ratio is chosen as 4% for fractal

summarization on handheld devices. On the other

hand, a threshold value is the maximum number of

sentences can be extracted from a range-block that is a

node in the document tree. If the quota allocated is

larger than the threshold value, the range-block must

be divided into subrange-block. Document summar-

ization is different from image compression, more

than one attractor can be chosen in one range-block.

The summarization by extraction of fixed number of

sentences is proven; the optimal length of summary is

3 to 5 sentences [16]. The default value of threshold is

chosen as 5 in our system.

The fractal value Fv of range-block r is computed

based on Range-block Significance Score (RBSS),

Document Fractal: 1 Quota: 40

Chapter 1 Fractal: 0.3 Quota: 12

Chapter 2 Fractal: 0.5 Quota: 20

Chapter 3 Fractal: 0.2 Quota: 8

Section 1.1 Fractal: 0.1 Quota: 4

Section 1.2 Fractal: 0.15

Quota: 6

Section 1.3 Fractal: 0.05

Quota: 2

Section 2.1 Fractal: 0.1 Quota: 3

Section 2.2 Fractal: 0.25 Quota: 10

Section 2.3 Fractal: 0.15

Quota: 7

Section 3.1 Fractal: 0.12

Quota: 5

Section 3.2 Fractal: 0.8 Quota: 3

Paragraphs ... Paragraphs... Paragraphs...

Fig. 8. An example of fractal summarization model.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 53

where the RBSS is computed as the sum of the

normalized weights of the sentence scores based on

the four salient features as introduced in Section 3.1.

Fv rð Þ ¼

1 if r is root

C Fv parent of rð Þ � RBSS rð ÞPRBSS xð Þ

xa sibling of r

0B@

1CA

1D

otherwise

8>>><>>>:

ð3Þ

The system utilizes the fractal value to compute the

sentence quota of each range-block by sharing the

quota of parent node directly proportional to the frac-

tal value of the child nodes, until the sentence quota

allocated is less than the threshold value, and the

system can extract sentences directly from the

range-block. During extraction of sentences in the

range-block, the system will extract the sentences

according to the fractalized sentence weight of the

sentences.

3.4. Experimental result

It is believed that a full-length text document con-

tains a set of subtopics [20] and a good quality

summary should cover as many subtopics as possible,

experiments showed that the fractal summarization

model produces a summary with a wider coverage

of information subtopic than traditional summariza-

tion model.

Experiment of fractal summarization and tradi-

tional summarization with four unweighted features

described in Section 3.1 was conducted on Hong

Kong Annual Report 2000. In the experiment, it is

found that the traditional summarization model

extracts most of sentences from few chapters. As

shown in Table 1, the traditional summarization

extracts 29 sentences from one chapter when the

sentence quota is 80 sentences, and 53 sentences are

extracted from top 3 chapters out of total 23 chapters,

no sentence is extracted from 8 chapters. However,

the fractal summarization extracts the sentences

evenly from each chapter. It extracts maximum 8

sentences and minimum 1 sentence from each chapter

(Table 1). The standard deviation of sentence number

extracted from chapters is 2.11 sentences in fractal

summarization against 6.55 sentences in traditional

summarization. Researchers believed that a good sum-

mary should find diverse topic areas in the text and

reduce the redundancy of information contents in the

summary [37]. Fractal summarization extracts the

sentences distributively from the document, therefore

it finds diverse topic areas and reduces the redundancy

of information at the same time.

A user evaluation is conducted. Ten subjects were

asked to evaluate the quality of summaries of 23

documents generated by fractal summarization and

traditional summarization. Both summaries of all

documents are assigned to each subject in random

order without telling the generation methods of the

summaries. The results show that all subjects consider

the summary generated by fractal summarization

method as a better summary. In order to compare

the result in more great detail, we calculate the preci-

sion as the number of relevant sentences in the sum-

mary accepted by the user divided by the number of

sentences in the summary (Table 2). The fractal sum-

marization can achieve up to 91.25% precision and

87.16% on average, while the traditional summariza-

tion can achieve up to a maximum of 77.50% preci-

sion and 67.00% on average. One-tailed T-test has

shown that the precision of fractal summarization

model outperforms the traditional summarization sig-

nificantly at 99% confidence level. Experiment of

fractal summarization and traditional summarization

with four unweighted features described in Section

3.1 was conducted on Hong Kong Annual Report

2000. In the experiment, it is found that the traditional

Table 1

Number of sentences extracted by two summarization models from

Hong Kong Annual Report 2000

Chapter

ID

Chapter

title

Number of

sentences

extracted in

fractal

summarization

model

Number of

sentences

extracted in

traditional

summarization

model

1 Hong Kong:

Asia’s World

City

6 3

2 Constitution and

Administration

4 1

3 The Legal System 2 0

4 The Economy 5 14

5 Financial and

Monetary affairs

8 29

6 Commerce and

Industry

6 10

7 Employment 2 2

8 Primary Production 1 0

9 Education 2 1

10 Health 1 0

11 Social Welfare 1 0

12 Housing 1 0

13 Land, Public

Works and Utilities

4 0

14 Transport 5 3

15 Infrastructure 1 0

16 The Environment 4 1

17 Travel and Tourism 1 1

18 Public Order 5 2

19 Communications,

the Media and

Information

Technology

6 6

20 Religion and

Custom

2 0

21 Recreation, Sport

and the Arts

5 3

22 Population and

Immigration

3 1

23 History 5 3

Table 2

Precision of summaries of Hong Kong Annual Report 2000 for two

summarization models

User ID Fractal

summarization

model (%)

Traditional

summarization

model (%)

User 1 81.25 71.25

User 2 85.00 67.50

User 3 80.00 56.25

User 4 85.00 63.75

User 5 88.75 77.50

User 6 81.25 61.25

User 7 91.25 76.25

User 8 86.25 58.75

User 9 85.00 65.00

User 10 87.50 72.50

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6154

summarization model extracts most of sentences from

few chapters. As shown in Table 1, the traditional

summarization extracts 29 sentences from one chapter

when the sentence quota is 80 sentences, and 53

sentences are extracted from top 3 chapters out of

total 23 chapters, no sentence is extracted from 8

chapters. However, the fractal summarization extracts

the sentences evenly from each chapter. It extracts a

maximum of 8 sentences and a minimum of 1 sen-

tence from each chapter (Table 1). The standard devia-

tion of sentence number extracted from chapters is

2.11 sentences in fractal summarization against 6.55

sentences in traditional summarization. Researchers

believed that a good summary should find diverse

topic areas in the text and reduce the redundancy of

information contents in the summary [37]. Fractal

summarization extracts the sentences evenly from

the document; therefore it finds diverse topic areas

and reduces the redundancy of information at the

same time.

We have also conducted another experiment using

the tested TIPSTER Text Summarization Evaluation

(SUMMAC) data. The TIPSTER Text Summarization

Evaluation (SUMMAC) is the first large-scale, devel-

oper-independent evaluation of automatic text sum-

marization systems [33]. The documents for the

TIPSTER evaluation are drawn from Text Retrieval

(TREC) [19] CDs 4 and 5. The categorization task in

SUMMAC focuses on generic summaries. The task

sought to find out whether a generic summary could

effectively present sufficient information to allow an

analyst to quickly and correctly categorize a docu-

ment. In order to compare the performance with other

summarization systems, we have conducted the cate-

gorization task to evaluate the performance of the

fractal summarization model. We have followed the

standard TIPSTER setting to conduct the classifica-

tion task using fractal summarization. 100 documents

are selected from the TIPSTER corpus, and 10% of

sentences are extracted from each document by fractal

summarization system. 15 subjects are involved in the

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 55

experiment. The subjects need to classify the docu-

ments based on the sentences extracted for each docu-

ment. The experiment result shows that the fractal

summarization system has a similar precision as

other summarization systems, and it has a better

recall. The fractal summarization achieves an F-

score [40] of 0.63 but other summarization systems

achieve a mean F-score of 0.42. Therefore, the fractal

summarization system outperforms other summariza-

tion systems. As the documents used are relatively

shorter in their length, most of them contain less than

100 sentences.

3.5. Visualization of fractal summarization on hand-

held devices

The summary generated by Fractal Summarization

Model is represented in a hierarchical tree structure.

The hierarchical structure of summary is suitable for

visualization of information on handheld devices.

However, a summary displayed in a small area of

handheld devices without visualization effect is still

difficult to read, it can be further enhanced by dis-

playing the sentences in different font sizes accord-

ing to their importance to help user to focus on

important information and search for information

easily.

WML is the markup language supported by wire-

less handheld devices. The basic unit of a WML file is

Fig. 9. Screen capture of WAP summarization system. (a) Hong Kong Ann

dCommunication, the Media and Information TechnologyT.

a deck; each deck must contain one or more cards.

The card element defines the content displayed to

users, and the card cannot be nested. Each card

links to another card within or across decks. Nodes

on the fractal tree of fractal summarization model are

converted into cards, and anchor links are utilized to

implement the tree structure. Given a card of a sum-

mary node, there may be a lot of sentences or child

nodes. A large number of sentences in a small display

area make it difficult to read. Fisheye View is a

visualization technique to enlarge the focus of interest

and diminish the information that is less important

[13]. When a user look at an object, the objects nearby

are shown in a larger visual size; and the visual size of

other objects is decreased inversely proportional to

their distances to the focus point.

In our system, we have modified the fisheye view.

The size of an object does not depend on its distance

from the focus point, but depends on the significance

of the object. The sentences are displayed in different

font sizes according to their importance. We have

implemented the fisheye view with 3-scale font

mode available for WML. The sentences or child

nodes are sorted by their sentence weights or fractal

value and separated evenly into three groups. The

group with highest value is displayed in bLargeQfont size, and the group with middle value and the

group with lowest value are displayed in bNormalQand bSmallQ font size, respectively.

ual Report 2000; (b) Chapter 19 of Hong Kong Annual Report 2000,

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6156

The prototype system using Nokia Handset Simu-

lator is presented in Fig. 9. The document presenting

is the Hong Kong Annual Report 2000. There are

totally 23 chapters in the annual report, 8 of them

are in large font, which means that they are more

important, and the rest are in normal font or small

font according to their importance to the report (Fig.

9a). The number inside the parentheses indicates the

number of sentences under the node that are extracted

as part of the summary. The main screen of the Hong

Kong Annual Report 2000 gives user a general idea of

overall information contents and the importance of

each chapter. If the user wants to explore a particular

node, user can click the anchor link, and the handheld

device sends the request to the WAP gateway, and the

gateway then decides whether to deliver another menu

or the summary of the node to the user depends on its

fractal value and quota allocated. Fig. 9b shows the

summary of Chapter 19 of Hong Kong Annual Report

2002, bCommunication, the Media and Information

TechnologyQ.A handheld PDA is usually equipped with more

memory, and the complete summary can be down-

loaded as a single WML file to the PDA through local

synchronization. To read the summary, the PDA is

required to install a standard WML file reader, i.e.,

KWML for Palm [24].

4. Summarization for financial news delivery on

handheld devices

Fractal summarization model summarizes the docu-

ments based on hierarchical document structure. In

addition to large text document, a lot of other informa-

tion sources also exhibit hierarchical document struc-

ture, such as web-site and newspaper. Due to the huge

information available, it is difficult to browse these

information sources on handheld devices. Automatic

summarization is a possible solution. Theoretically, the

fractal summarization is capable to summarize to all

these information sources as long as the calculation of

fractal value is well formulated. As financial news is

critical in decision making, we shall modify the fractal

value formula of generic fractal summarization in order

to summaries the financial news and we shall demon-

strate financial news delivery with fractal summariza-

tion on handheld devices.

4.1. Fractal summarization of financial news

Fractal summarization model performs the sum-

marization based on hierarchical document struc-

ture. In addition to large text documents, a lot of

other documents also exhibit hierarchical tree docu-

ment structure, such as web-site, newspaper, etc.

The fractal summarization model is capable to

summarize these documents based on their structure

and their relationships in categorization; therefore, it

is a powerful tool in providing m-services of real

time information delivery. As present, a lot of

electronic news delivery services have been pro-

vided. An example of the fractal summarization

model being used to summarize the financial

news available from Internet is presented in this

section.

Newspaper is one of the documents that exhibit

the well-defined hierarchical document structure. At

present, there is a lot of electronic news delivery

services provided for PC, and most of them provide

summarization tools to help user to search informa-

tion, such as Lycos Financial Feed System with

summarization system from Diyatech [9], Yellow-

Brix with Inxight’s Summarizer [52] and Columbia’s

Newsblaster [34]. However, summarizers for PC

platform are not adaptable to mobile devices directly.

Moreover, the existing commercial summarizers are

indeed extracting the first few sentences from the

document or using the primitive summarization

model without considering the hierarchical structure

of documents or the organization of information.

Yahoo!News [47] is one of the most popular online

content providers. There are 21 categories in the

Yahoo!News. Moreover, each of the categories will

be sub-divided into subcategories. Take the Business

category as an example (Fig. 10), this category con-

tains financial news and it is sub-divided into six

sub-categories, namely, Economy, Stock Markets,

Earnings, Personal Finance, Industries and Commen-

tary. Each sub-category contains around 10 news

articles. Each news article is a tree structure by itself.

For some longer news article, there may exist more

than one section, and each section contains few

paragraphs, and paragraph contains sentences.

The Fractal Summarization of Yahoo!News is very

similar to the fractal summarization of large text

document, only some minor modifications are

Yahoo! News - BusinessWeight: 1 Quota: 40

Commentary Weight: 0.25

Quota: 10

Earning Weight: 0.25

Quota: 10

Economy Weight: 0.2

Quota: 8

Industries Weights: 0.1

Quota: 4

Personal Finance Weight: 0.15

Quota: 6

Stock Market Weight: 0.05

Quota: 2

Free Flight Weight: 0.0075

Quota: 3

News ... News ... News ... News ...

Fig. 10. Fractal summarization of yahoo! news-dbusinessT category.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 57

required to demonstrate the characteristic of the

Yahoo!News.

! Firstly, the headings of categories and subcate-

gories do not have a direct impact on the content

of news under the branch; it serves for classifica-

tion purpose only. As a result, the heading method

will consider the headings of news articles only. In

addition, the headings of categories can be used for

personalization of news delivery, the user can set

his preference of each category in advance and the

system will adjust the weights accordingly. Alter-

natively, the preference can be constructed by auto-

learning of machine in the middle-tier. The WAP

gateway can analyze the reading behavior of user

and predict the user’s preference.

! The location feature in traditional summarization

assumes that the text unit in the beginning or ending

is more important. The news articles inside a sub-

category are sorted in chronological order. The most

recent news is usually considered as more impor-

tant. Therefore, we propose calculating the location

weight of a news article by its chronological position

in the subcategory or the time-lag between the news

event and browsing time. However, when the sys-

tem traces the summarization tree down to a node

inside a news article, the generic location method in

fractal summarization will be adopted.

! In order to provide a glimpse of every article, each

news article will receive a sentence quota with at

least one sentence.

4.2. Financial news delivery to handheld devices

In order to minimize the bandwidth requirement

and reduce the pressure on computing power of hand-

held devices, the summarization of Yahoo!News will

be conducted in two levels. As high-compression ratio

abstracting is more useful [43] and it can save the

network bandwidth, the fractal summarization system

generates a brief skeleton of summary with compres-

sion ratio equals to 4% at the first stage. The details of

the summary at different levels of the news tree are

generated on demands of users.

When handheld device retrieves the financial news

fromYahoo! News-Business, the systemwill first show

a card containing with 6 subcategories of dBusinessTcategory (Fig. 11). In the figure, three subcategories of

Business category are displayed in large font, which

means that they are more important; and the rest are in

normal font or small font according to their importance.

The skeleton of news gives user a general idea how the

news articles are organized, and the user can decide

which subcategory to go into details. When the user

clicks the anchor link of subcategory, theWAP gateway

will deliver a card depends on the quota allocated. If a

large quota is allocated to the subcategory, the system

will show another card containing of index of news

article. However, if the quota is less than a threshold

value of 5 sentences, the system will show a card with

the summary of all news articles in the subcategory. In

the summary page, when the user clicks the anchor link

dMoreT at end of sentences, the systemwill generate the

summary for the corresponding news articles with

compression ratio 20%, because it has been proved

that extraction of 20% sentences can be as informative

as the full text of the source document [36]. On the

other hand, the user can click the anchor link dFullT toview the full text of the news articles. Such interactive

summarization reduces the computation load in com-

paring with the generation of the entire summary in one

batch by the traditional automatic summarization,

which is ideal for m-services.

4.3. Future work

In current stage, the fractal summarization is capable

to process textual information only. However, there is a

Fig. 11. Financial news delivery system on mobile devices.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6158

lot of information available inmultimedia format on the

Web. Information delivery of multimedia document

will be one of the key research topics in the near future.

As the multimedia documents require a much higher

bandwidth than textual documents, this problem cannot

be resolved solely by the current steaming technology.

Summarization of multimedia documents is required

for information delivery to mobile devices. The

research work of spoken document was initiated in

the spoken language track of TREC 1997 [14]. The

research of spoken document summarization starts in

2000 [53,54]. Summarization of video has also been

investigated [45,46]. It would be a great challenge to

move the proposed model to multimedia documents.

The summarization of multimedia document is a com-

plementary to the proposed model. Nowadays, most of

the mobile devices are speech-based. With the sum-

marization of spoken documents, the information can

be easily delivered to speech-based mobile devices.

This will certainly increase the popularity of the pro-

posed model. Moreover, it can provide information

access for the blind as well [17].

5. Conclusion

Mobile commerce is a promising addition to the

electronic commerce by the adoption of portable hand-

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 59

held devices. However, the mobile computing should

not be limited to user-centered m-services applications

only, it should be extended to decision making in an

organization. With a fast paced economy, organization

need to make a decision as fast as possible, access to

large text documents or other information sources is

important during decisionmaking. Unfortunately, there

are many shortcomings of the handheld devices, such

as limited resolution and narrow bandwidth. In order to

overcome the shortcomings, fractal summarization and

information visualization are proposed in this paper,

which are critical in decision support in an m-organiza-

tion. The fractal summarization creates a summary in

hierarchical tree structure and presents the summary to

the handheld devices through cards in WML. The

adoption of keyword feature, location feature, heading

feature, and cue feature are discussed. Users may

browse the selected summary by clicking the anchor

links from the highest abstraction level to the lowest

abstraction level. Fractal views are utilized to filter the

less important nodes in the document structure, sen-

tences are displayed in different font size to enlarge the

focus of interest and diminish the less significant sen-

tences. Such visualization effect draws users’ attention

on the important content. The three-tier architecture is

presented to reduce the computing load of the handheld

devices. The proposed system creates an information

visualization environment to avoid the existing short-

comings of handheld devices for mobile commerce.

References

[1] M.F. Barnsley, A.E. Jacquin, Application of recurrent iterated

function systems to images, Proceedings of SPIE Visual Com-

munications and Image Processing (VCIP’88), Cambridge,

MA, USA, vol. 1001, SPIE, 1988 (Nov.), pp. 122–131.

[2] P.B. Baxendale, Machine-made index for technical literature—

an experiment, IBM Journal of Research and Development 2

(4) (1958 (Oct.)) 354–361.

[3] B. Boguraev, R. Bellamy, C. Swart, Summarization miniatur-

ization: delivery of news to handhelds, Proceedings of Work-

shop on Automatic Summarization 2001, pp. 99–110, in

conjunction with The Second Meeting of the North American

Chapter of the Association for Computational Linguistics

(NAACL 2001), Association for Computational Linguistics,

Pittsburgh, PA, USA, 2001 (Jun.).

[4] M.H. Brown, W.E. Weihl, Zippers: a Focus+Context display

of web pages, Proceedings of World Conference of the Web

Society (WebNetT96), San Francisco, CA, USA 1996 (Oct.),

DEC SRC Technical Report 140, (1996 (May)).

[5] O. Buyukkokten, H. Garcia-Molina, A. Paepcke, T. Winograd,

Power browser: efficient web browsing for PDAs, Proceedings

of the SIGCHI Conference on Human Factors in Computing

Systems (CHI 2000), ACM Press, Hague, Netherlands, 2000

(Apr.), pp. 430–437.

[6] O. Buyukkokten, H. Garcia-Molina, A. Paepcke, Seeing the

whole in parts: text summarization for web browsing on

handheld devices, Proceedings of the 10th International Con-

ference on World Wide Web (WWW10), ACM Press, Hong

Kong, China, 2000 (May), pp. 652–662.

[7] O. Buyukkokten, H. Garcia-Molina, A. Paepcke, Accordion

summarization for end-game browsing on PDAs and cellular

phones, Proceedings of the SIGCHI Conference on Human

Factors in Computing System (CHI 2001), ACM Press, Seat-

tle, WA, USA, 2001 (Mar.), pp. 213–220.

[8] O. Buyukkokten, H. Garcia-Molina, A. Paepcke, Text sum-

marization of web pages on handheld devices, Proceedings of

Workshop on Automatic Summarization 2001, in conjunction

with The Second Meeting of the North American Chapter of

the Association for Computational Linguistics (NAACL

2001), Association for Computational Linguistics, Pittsburgh,

PA, USA, 2001 (Jun.).

[9] Diyatech Homepage, http://www.diyatech.com/clycos.htm.

[10] H.P. Edmundson, New method in automatic extraction, Jour-

nal of the ACM 16 (2) (1969 (Apr.)) 264–285.

[11] B. Endres-Niggemeyer, E. Maier, A. Sigel, How to implement

a naturalistic model of abstracting: four core working steps of

an expert abstractor, Information Processing and Management

31 (5) (1995 (Sep.)) 631–674.

[12] J. Feder, Fractals, Plenum, New York, 1988.

[13] G.W. Furnas, Generalized fisheye views, Proceedings of the

SIGCHI Conference on Human Factors in Computing Sys-

tem (CHI ’86), ACM SIGCHI Bulletin, vol. 17 (4), 1986

(Apr.), pp. 16–23.

[14] J.S. Garofolo, E.M. Voorhess, V.M. Stanford, K.S. Jones,

TERC-6 1997 spoken document retrieval track overview and

results, Proceeding of the Sixth Text REtrieval Conference

(TREC 6), National Institute of Standards and Technology

(NIST), Gaithersburg, MD, USA, 1997 (Nov.), pp. 83–91.

[15] B.G. Glaser, A.L. Strauss, The Discovery of Grounded The-

ory; Strategies for Qualitative Research, Aldine de Gruyter,

New York, 1967.

[16] J. Goldstein, M. Kantrowitz, V. Mittal, J. Carbonell, Summar-

izing text documents: sentence selection and evaluation

metrics, Proceedings of the Twenty-Second Annual Interna-

tional ACM-SIGIR Conference on Research and Development

in Information Retrieval, (SIGIR’99), Berkeley, California,

USA, ACM Press, 1999 (Aug.), pp. 121–128.

[17] G. Grefenstette, Producing intelligent telegraphic text reduc-

tion to provide an audio scanning service for the blind, Work-

ing Notes of the American Association for Artificial

Intelligence 1998 Spring Symposium (AAAI’98) on Intelli-

gent Text Summarization, Stanford, California, AAAI Press,

1998 (Mar.), pp. 111–117.

[18] D.K. Harman, Ranking algorithms, in: W.B. Frakes, R. Baeza-

Yates (Eds.), Information Retrieval: Data Structures and Algo-

rithms, Prentice-Hall, 1992, pp. 363–392 (Ch. 14).

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–6160

[19] D.K. Harman, E.M. Voorhees, The Fifth Text REtrieval Con-

ference (TREC-5), NIST Special Publication, vol. 500-238,

National Institute of Standards and Technology, Gaithersburg,

MD, USA, 1996 (Nov.).

[20] M.A. Hearst, Subtopic structuring for full-length document

access, Proceedings of the 16th Annual International ACM

SIGIR Conference on Research and Development in Informa-

tion Retrieval (SIGIR’93), Pittsburgh, Pennsylvania, USA,

1993 (Jun.), pp. 56–68.

[21] A.E. Jacquin, Fractal image coding: a review, Proceedings of

the IEEE, vol. 81 (10), 1993 (Oct.), pp. 1451–1465.

[22] J. Kepiec, J. Pedersen, F. Chen, A trainable document sum-

marizer, Proceedings of the 18th Annual International ACM

SIGIR Conference on Research and Development in Informa-

tion Retrieval (SIGIR’95), Seattle, Washington, USA, 1995

(Jul.), pp. 68–73.

[23] H. Koike, Fractal views: a fractal-based method for controlling

information display, ACM Transactions on Information Sys-

tems (TOIS) 13 (3) (1995 (Jul.)) 305–323.

[24] KWML. KWML-KVM WML (WAP) Browser on Palm.

PALM Inc., http://www.jshape.com/kwml/index.html, (2002).

[25] M. Lam-Adesina, G.J.F. Jones, Applying summarization tech-

niques for term selection in relevance Feedback, Proceedings

of the 24th Annual International ACM SIGIR Conference on

Research and Development in Information Retrieval

(SIGIR’01), New Orleans, Louisiana, USA, ACM Press,

2001 (Sep.), pp. 1–9.

[26] M. Lamming, M. Eldridge, M. Flynn, C. Jones, D. Pendlebury,

Satchel: providing access to any document, any time, any-

where, ACM Transactions on Computer-Human Interaction

(TOCHI), special issue on human–computer interaction with

mobile systems 7 (3) (2000 (Sep.)) 322–352.

[27] Y. Lin, E.H. Hovy, Identifying topics by position, Proceed-

ings of the Workshop of Intelligent Scalable Text Summari-

zation, in conjunction of the Fifth Conference on Applied

Natural Language Processing (ANLP-97), Washington, DC,

Association for Computational Linguistics, 1997 (Mar.),

pp. 283–290.

[28] H.P. Luhn, The automatic creation of literature abstracts, IBM

Journal of Research and Development 2 (2) (1958 (Apr.))

159–165.

[29] I.S. MacKenzie, R. Soukoreff, Text entry for mobile comput-

ing: models and methods, theory and practice, Human–Com-

puter Interaction 17 (2 and 3) (2002) 147–198.

[30] I.S. MacKenzie, Mobile text entry using three keys, Proceed-

ings of the Second Nordic Conference on Human Computer

Interaction. (NordiCHI 2002), ACM Press, Aarhus, Denmark,

2002 (Oct.), pp. 27–34.

[31] B. Mandelbrot, The Fractal Geometry of Nature, W.H. Free-

man, New York, 1983.

[32] I. Mani, Recent development in text summarization, The

Proceedings of the tenth International Conference on Informa-

tion and Knowledge Management (CIKM’01), ACM Press,

Atlanta, GA, USA, 2001 (Nov.), pp. 529–531.

[33] I. Mani, D. House, G. Klein, L. Hirschman, T. Firmin, B.

Sundheim, The TIPSTER SUMMAC text summarization eva-

luation, Proceedings of the Ninth Conference on European

Chapter of the Association for Computational Linguistics

(EACL’99), University of Bergen, Bergen, Norway, 1999

(Jun.), pp. 77–85.

[34] K. McKeown, R. Barzilay, J. Chen, D. Elson, D. Evans, J.

Klavans, A. Nenkova, B. Schiffman, S. Sigelman, Columbia’s

newsblaster: new features and future directions, Proceedings

of the 2003 Human Language Technology conference of the

North American Chapter of the Association for Computational

Linguistics (HLT-NAACL 2003), Demonstrations, Associa-

tion for Computational Linguistics, Edmonton, Canada, 2003

(May), pp. 15–16.

[35] T. Mendelsohn, Content at the point-of-care: in the palm of a

doctor’s hand, Proceedings of Online Information, London,

U.K., 2001 (Dec.), pp. 169–174.

[36] A.H. Morris, G.M. Kasper, D.A. Adams, The effects and

limitations of automated text condensing on reading compre-

hension performance, Information Systems Research 3 (1)

(1992 (Mar.)) 17–35.

[37] T. Nomoto, Y. Matsumoto, A new approach to unsupervised

text summarization, Proceedings of the 24th Annual Interna-

tional ACM SIGIR Conference on Research and Development

in Information Retrieval (SIGIR’01), ACM Press, New

Orleans, LA, USA, 2001 (Sep.), pp. 26–34.

[38] PALM. PALM: Providing Fluid Connectivity in a Wireless

World. White Paper of Palm Inc., http://www.palm.com/

wireless/ProvidingFluidConnectivity.pdf, (2002).

[39] K. Perlin, Quikwriting: continuous stylus-based text entry,

Proceedings of the 11th Annual ACM Symposium on User

Interface Software and Technology (UIST’98), ACM Press,

San Francisco, CA, USA, 1998 (Nov.), pp. 215–216.

[40] C.J. van Rijsbergen, Information Retrieval, Butterworths, Lon-

don, 1979.

[41] G. Salton, C. Buckley, Term-weighting approaches in auto-

matic text retrieval, Information Processing and Management

24 (5) (1998 (May)) 513–523.

[42] Y. Seki, K. Eguchi, N. Kando, Compact Summarization for

Mobile Phones, Proceedings of Mobile HCI 2003 International

Workshop, Udine, Italy, 2003 (Sep.), In: F. Crestani, M.

Dunlop, S. Mizzaro, (Eds.), Mobile and Ubiquitous Informa-

tion Access, Lecture Notes in Computer Science, 2954, pp.

172–196, Springer, 2004.

[43] S. Teufel, M. Moens, Sentence extraction and rhetorical clas-

sification for flexible abstracts, Proceedings of the 1998 AAAI

Spring Symposium on Intelligent Text Summarization, AAAI

Press, Palo Alto, CA, USA, 1998 (Mar.), pp. 16–25.

[44] S. Teufel, M. Moens, Sentence extraction as a classification

task, Proceedings of the ACL’97/EACL’97 Workshop on

Intelligent Scalable Text Summarization, Universidad Nacio-

nal de Educacion a Distancia (UNED), Madrid, Spain.

Morgan Kaufmann Publishers, Madrid, Spain, 1997 (Jul.),

pp. 58–68.

[45] N. Vasconcelos, A. Lippman, Bayesian modeling of video

editing and structure: semantic features for video summar-

ization and browsing, Proceedings of 1998 IEEE Interna-

tional Conference on Image Processing (ICIP’98), vol. 3,

IEEE Computer Society, Chicago, IL, USA, 1998 (Oct.),

pp. 153–157.

C.C. Yang, F.L. Wang / Decision Support Systems 43 (2007) 46–61 61

[46] N. Vasconcelos, A. Lippman, A spatiotemporal motion model

for video summarization, Proceedings of the IEEE Computer

Society Conference on computer Vision and Pattern Recogni-

tion (CVPR’98), IEEE Computer Society, Santa Barbara, CA,

USA, 1998 (Jun.), pp. 361–366.

[47] Yahoo!News, 2003, Yahoo!News homepage, http://news.

yahoo.com.

[48] C.C. Yang, F.L. Wang, Fractal summarization: summarization

based on fractal theory, Proceedings of the 26th Annual Inter-

national ACM SIGIR Conference: Research and Development

in Information Retrieval (SIGIR 2003), ACM Press, Toronto,

Canada, 2003 (Jul.).

[49] C.C. Yang, F.L. Wang, Fractal summarization for mobile

devices to access large documents on the web, Proceedings

of the Twelfth International Conference on World Wide Web

(WWW 2003), ACM Press, Budapest, Hungary, 2003 (May),

pp. 215–224.

[50] C.C. Yang, F.L. Wang, Automatic summarization for financial

news delivery on mobile devices, Proceedings of the Twelfth

International Conference on World Wide Web (WWW 2003),

ACM Press, Budapest, Hungary, 2003 (May), pp. 391–392.

[51] C.C. Yang, F.L. Wang, Document summarization on handheld

device: an information visualization tool for mobile com-

merce, Proceeding of The First Workshop on e-Business

(WEB2002) in International Conference on Information Sys-

tems (ICIS 2002), Association for Information Systems, Bar-

celona, Spain, 2002 (Dec.).

[52] YellowBrix, 2003, YellowBrix Homepage, http://www.

yellowbrix.com/.

[53] K. Zechner, A. Waibel, DiaSumm: flexible summarization of

spontaneous dialogues in unrestricted domains, Proceedings of

the 18th International Conference on Computational Linguis-

tics (COLING-2000), Saarbruecken, Germany, Morgan Kauf-

mann Publishers, Universitat des Saarlandes, Saarbrucken,

Germany, 2000 (Jul.), pp. 968–974.

[54] K. Zechner, A. Waibel, Minimizing word error rate in textual

summaries of spoken language, Proceedings of the 1st Con-

ference of the North American Chapter of the Association for

computational Linguistics and the 6th Conference on Applied

Natural Language Processing (NAACL/ANLP 2000), Asso-

ciation for Computational Linguistics, Seattle, WA, USA,

2000 (Apr.), pp. 186–193.

Christopher C. Yang is an associate professor in the Department of

Systems Engineering and Engineering Management at the Chinese

University of Hong Kong. He received his BS, MS, and PhD in

Electrical and Computer Engineering from the University of Ari-

zona. Before he joined the Chinese University of Hong Kong, he

was an assistant professor in the Department of Computer Science

and Information Systems and the associate director of the Author-

ized Academic JavaSM CampusSM at the University of Hong Kong.

He has also been a research scientist in the Artificial Intelligence

Laboratory in the Department of Management Information Systems

at the University of Arizona. His recent research interests include

cross-lingual information retrieval, multimedia information retrie-

val, digital library, information visualization, Internet searching,

automatic summarization, information behavior, and electronic

commerce. He has published over 100 refereed journal and con-

ference papers in Journal of the American Society for Information

Science and Technology, IEEE Transactions on Image Processing,

IEEE Transactions on Robotics and Automation, IEEE Computer,

Information Processing and Management, Decision Support Sys-

tems, Graphical Models and Image Processing, Optical Engineering,

Pattern Recognition, International Journal of Electronic Commerce,

Applied Artificial Intelligence, SIGIR, ICIS, WWW, and more. He

was the chairman of the Association for Computing Machinery

Hong Kong Chapter, the program co-chair of the First International

Conference on Asia Digital Library, and a program committee and

organizing committee member for several international conferences

in digital library, information systems, electronic commerce, com-

puter vision, image processing, and information retrieval. He has

frequently served as an invited panelist in the NSF Digital Library

Initiative Review Panel and the NSF Information Technology

Research Review Panel in US.

Fu Lee Wang is an Instructor in the Department of Computer

Science, City University of Hong Kong. He received B.Eng. (Hon)

degree in Computer Engineering and M.Phil. degree in Computer

Science and Information Systems from the University of Hong Kong,

and PhD degree in Systems Engineering and Engineering Manage-

ment from Chinese University of Hong Kong. His current research

interests focus on document summarization, digital library, and

information retrieval. His research work has been appeared in such

journals as Journal of the American Society for Information Science

and Technology, and Information Processing Letters.