COMMUNICATIONS OF THE ACM February 2007/Vol. 50, No. 2 75

By Dongsong Zhang

e have witnessed the explosive growth of mobilehandheld devices such as cell phones and personal

digital assistants (PDAs) in recent years. Manywireless applications have been developed for those

devices, including daily news update, classifiedadvertising, tourist guide, wireless Web portals, and

m-commerce applications. The ability to communi-cate from virtually anywhere and the convergence of

Web and wireless technologies offer an unprecedentedlevel of flexibility and convenience, particularly for ubiq-

uitous information access through mobile devices. How-ever, the unique features of wireless networks (for example,

low bandwidth and unreliability) and mobile devices (forexample, small screen size, and low memory and processing capability), as well as themobility of users, present challenges for taking advantage of the convenience ofmobile devices for information access. For example, most Web content is designed

WCreative technologies can help overcome the limitations of mobile

handheld devices while maximizing their strengths.

Web ContentAdaptationfor Mobile Handheld Devices

based on the user’s previous access history, while theprefetch thresholds computed based on system andnetwork conditions, cost of bandwidth, and trans-mission time determine whether the performancemay be improved by prefetching certain content.

The content delivery can also be context-aware.The term context refers to information that charac-terizes a situation related to the interaction betweenusers, mobile applications, and the surrounding envi-

ronment, such as informationabout a person, a place, or anobject. Context information gath-ered from various sensors, net-works, devices, user profiles, and other sources cantrigger the content adaptation in order to cope withthe dynamic environment. For example, in a museum,mobile device users should only receive correspondingintroductory information when they enter differentexhibition rooms. In the GUIDE project (a context-aware electronic tour guide) [4], visitors to a city canaccess context-aware information and services usingtheir handheld GUIDE units. The information pre-sented to visitors is tailored based on the visitor’s inter-est, current physical location, and attractions alreadyvisited.

ADAPTIVE CONTENT PRESENTATION

While many of today’s mobile devices already featureWeb browsers, browsing the Web on a mobile devicehas not become as convenient as expected. Mobilehandheld devices are quite restrictive on the formatand length of the received content. They typicallydisplay less than 20 lines of text on the screen; theymay run different operating systems and support dif-ferent markup languages. There is information losswhen a Web site uses a presentational mode that amobile device does not support. Currently, mostWeb pages are designed only for display on desktopcomputers, making direct presentation of thosepages on small devices aesthetically unpleasant, diffi-

cult to navigate, and even completely illegible. Merely squeezing original Web content into small

screens may not work. Traditionally, device-specificauthoring and multiple-device authoring approacheswere used to adapt Web content presentation foreffective display on mobile devices. They eitherauthorized a set of Web pages that were designed andformatted up-front according to a predefined guide-line for a specific device, or generated a set of differ-

ent versions of the same Web page to cover a numberof identified target devices. Obviously, thoseapproaches are device dependent and inflexible.Therefore, automatic re-authoring and client-sidenavigation approaches seem more effective. Suchapproaches involve developing software to re-autho-rize a Web page in real time through a series of trans-formations, including layout change and contentformat reconfigurations, so that the page can beeffectively displayed on a device and the user caninteractively navigate the page.

Similar to a user profile, a device profile can beused to support device-related adaptation. It specifiesthe MIME media types and physical characteristics ofa device including color depth, screen size, memorysize, operating system, as well as supported markuplanguages. The generic Composite Capabilities/Pref-erence Profiles (CC/PP) framework (www.w3.org/2001/di/) provides a mechanism throughwhich a mobile user agent, such as a browser, cantransmit information about the mobile device. A useragent profile based on the CC/PP framework includesdevice hardware and software characteristics, informa-tion about the network to which the device is con-nected, and other attributes [6].

Figure 1 shows a few screen shots of a Yahoo! Webpage displayed on a HP iPAQ h4355 Pocket PC witha 3.5” transflective screen (Figure 1(a)). It has inte-grated Bluetooth and WLAN 802.11b for wirelesscommunication and access to the Internet. Figure1(b) shows the page displayed on that device without

COMMUNICATIONS OF THE ACM February 2007/Vol. 50, No. 2 7776 February 2007/Vol. 50, No. 2 COMMUNICATIONS OF THE ACM

and optimized for desktop and broadband clients.Web content is poorly suited for mobile devices[11]; users who carry those devices around may needdifferent information under different contexts.Therefore, how to adapt content to meet the needsof users, fit the characteristics of individual mobiledevices, and adjust to dynamic context becomesimportant.

The World Wide Web Consortium (W3C) definescontent adaptation as a process of selection, genera-tion, or modification that produces one or more per-ceivable units in response to arequested uniform resource. In2005, W3C announced the launchof a new Mobile Web Initiative(MWI), aiming to address the inter-operability and usability problems inorder to make “Web access from amobile device as simple, easy, andconvenient as Web access from adesktop device.” This article dis-cusses Web content adaptation formobile devices from several perspec-tives.

SELECTIVE CONTENT DELIVERY

There are two types of content delivery to mobiledevice users: pull and push. In the pull model, amobile client sends an information query to a serverand the server returns the relevant content back. Inthe push model, a server automatically delivers con-tent to mobile clients via a push proxy withoutreceiving users’ requests. The push model is veryuseful in mobile applications such as news alert ser-vices, mobile advertising, and real-time trafficupdates.

To prevent users from wading through a morass ofpossibly irrelevant content to find a single gem ofinformation of interest, selecting and delivering con-tent relevant to users’ interest is essential. This can beachieved by employing user profiles. A push proxyassesses the similarity between a Web page and a userprofile to determine if it might be of interest to theuser. Furthermore, a Web page can be partitionedinto several blocks (for example, sports, business,health, advertisement, among others). Unnecessary orirrelevant blocks can be removed from the originalpage. A user profile in a mobile information systemcan consist of the following information about a user’sinterests and preferences:

• Demographic information;• Information interests (for example, represented

by keywords);

• Browsing history, such as recently visited Websites, the time of last access, and visiting frequencies;

• Content presentation preferences, which will bediscussed later;

• Quality of Service (QoS) preferences; and• Access privilege indicating what information a

user can access;

User interests can be inferred and updated based onusers’ explicit and/or implicit feedback (for example,

browsing behavior on mobile devices) [2]. In amobile computing environment, user profiles can bestored at different locations, including a single cen-tralized server as the only profile server; differentprofile servers with duplicated or unduplicated userprofiles; or local mobile devices.

The push model is generally used for multicastinginformation to a group of users. It, coupled with userprofiles and intelligent agents, can also be used forcontent prefetching for individual users to minimizethe transmission delay (sometimes the required infor-mation is not even accessible when it is needed dueto a disconnected wireless network). The idea ofprefetching is that a system delivers certain contentto a local mobile device that will likely be accessed bythe user soon. However, the prefetching decision iscontingent upon different network conditions andthe likelihood that the user will access this contentshortly. Prefetching operations consume the alreadylimited bandwidth of wireless networks and limitedstorage of mobile devices, and users often must payfor the service. It will be particularly expensive if theuser does not view the prefetched content eventually.Therefore, it is vital not only to determine what con-tent should be prefetched and when, but also tobuild utility models to analyze the trade-off betweenthe potential benefit of prefetching and cost undercurrent circumstances (for example, network condi-tions). For instance, [5] proposes a prefetching deci-sion scheme that utilizes the access probabilities andprefetch thresholds. The access probabilities indicatehow likely certain content will be requested by a user

Figure 1. Adaptivedisplay of Web

content with reconfigurations.

Figure 2. A DOM-treenavigation of the MSNhome page via Mobile

Web.

Figure 1(a) Figure 1(b) Figure 1(c) Figure 1(d)

Figure 1. Adaptive Display of Web Content with Reconfigurations Zhang fig 1 (2/07)

Figure 1. Adaptive Display of Web Content with Reconfiguration.

Figure 1 (a) Figure 1 (b) Figure 1 (c) Figure 1 (d)(a) (b) (c) (d)

based on the user’s previous access history, while theprefetch thresholds computed based on system andnetwork conditions, cost of bandwidth, and trans-mission time determine whether the performancemay be improved by prefetching certain content.

The content delivery can also be context-aware.The term context refers to information that charac-terizes a situation related to the interaction betweenusers, mobile applications, and the surrounding envi-

ronment, such as informationabout a person, a place, or anobject. Context information gath-ered from various sensors, net-works, devices, user profiles, and other sources cantrigger the content adaptation in order to cope withthe dynamic environment. For example, in a museum,mobile device users should only receive correspondingintroductory information when they enter differentexhibition rooms. In the GUIDE project (a context-aware electronic tour guide) [4], visitors to a city canaccess context-aware information and services usingtheir handheld GUIDE units. The information pre-sented to visitors is tailored based on the visitor’s inter-est, current physical location, and attractions alreadyvisited.

ADAPTIVE CONTENT PRESENTATION

While many of today’s mobile devices already featureWeb browsers, browsing the Web on a mobile devicehas not become as convenient as expected. Mobilehandheld devices are quite restrictive on the formatand length of the received content. They typicallydisplay less than 20 lines of text on the screen; theymay run different operating systems and support dif-ferent markup languages. There is information losswhen a Web site uses a presentational mode that amobile device does not support. Currently, mostWeb pages are designed only for display on desktopcomputers, making direct presentation of thosepages on small devices aesthetically unpleasant, diffi-

cult to navigate, and even completely illegible. Merely squeezing original Web content into small

screens may not work. Traditionally, device-specificauthoring and multiple-device authoring approacheswere used to adapt Web content presentation foreffective display on mobile devices. They eitherauthorized a set of Web pages that were designed andformatted up-front according to a predefined guide-line for a specific device, or generated a set of differ-

ent versions of the same Web page to cover a numberof identified target devices. Obviously, thoseapproaches are device dependent and inflexible.Therefore, automatic re-authoring and client-sidenavigation approaches seem more effective. Suchapproaches involve developing software to re-autho-rize a Web page in real time through a series of trans-formations, including layout change and contentformat reconfigurations, so that the page can beeffectively displayed on a device and the user caninteractively navigate the page.

Similar to a user profile, a device profile can beused to support device-related adaptation. It specifiesthe MIME media types and physical characteristics ofa device including color depth, screen size, memorysize, operating system, as well as supported markuplanguages. The generic Composite Capabilities/Pref-erence Profiles (CC/PP) framework (www.w3.org/2001/di/) provides a mechanism throughwhich a mobile user agent, such as a browser, cantransmit information about the mobile device. A useragent profile based on the CC/PP framework includesdevice hardware and software characteristics, informa-tion about the network to which the device is con-nected, and other attributes [6].

Figure 1 shows a few screen shots of a Yahoo! Webpage displayed on a HP iPAQ h4355 Pocket PC witha 3.5” transflective screen (Figure 1(a)). It has inte-grated Bluetooth and WLAN 802.11b for wirelesscommunication and access to the Internet. Figure1(b) shows the page displayed on that device without

COMMUNICATIONS OF THE ACM February 2007/Vol. 50, No. 2 7776 February 2007/Vol. 50, No. 2 COMMUNICATIONS OF THE ACM

and optimized for desktop and broadband clients.Web content is poorly suited for mobile devices[11]; users who carry those devices around may needdifferent information under different contexts.Therefore, how to adapt content to meet the needsof users, fit the characteristics of individual mobiledevices, and adjust to dynamic context becomesimportant.

The World Wide Web Consortium (W3C) definescontent adaptation as a process of selection, genera-tion, or modification that produces one or more per-ceivable units in response to arequested uniform resource. In2005, W3C announced the launchof a new Mobile Web Initiative(MWI), aiming to address the inter-operability and usability problems inorder to make “Web access from amobile device as simple, easy, andconvenient as Web access from adesktop device.” This article dis-cusses Web content adaptation formobile devices from several perspec-tives.

SELECTIVE CONTENT DELIVERY

There are two types of content delivery to mobiledevice users: pull and push. In the pull model, amobile client sends an information query to a serverand the server returns the relevant content back. Inthe push model, a server automatically delivers con-tent to mobile clients via a push proxy withoutreceiving users’ requests. The push model is veryuseful in mobile applications such as news alert ser-vices, mobile advertising, and real-time trafficupdates.

To prevent users from wading through a morass ofpossibly irrelevant content to find a single gem ofinformation of interest, selecting and delivering con-tent relevant to users’ interest is essential. This can beachieved by employing user profiles. A push proxyassesses the similarity between a Web page and a userprofile to determine if it might be of interest to theuser. Furthermore, a Web page can be partitionedinto several blocks (for example, sports, business,health, advertisement, among others). Unnecessary orirrelevant blocks can be removed from the originalpage. A user profile in a mobile information systemcan consist of the following information about a user’sinterests and preferences:

• Demographic information;• Information interests (for example, represented

by keywords);

• Browsing history, such as recently visited Websites, the time of last access, and visiting frequencies;

• Content presentation preferences, which will bediscussed later;

• Quality of Service (QoS) preferences; and• Access privilege indicating what information a

user can access;

User interests can be inferred and updated based onusers’ explicit and/or implicit feedback (for example,

browsing behavior on mobile devices) [2]. In amobile computing environment, user profiles can bestored at different locations, including a single cen-tralized server as the only profile server; differentprofile servers with duplicated or unduplicated userprofiles; or local mobile devices.

The push model is generally used for multicastinginformation to a group of users. It, coupled with userprofiles and intelligent agents, can also be used forcontent prefetching for individual users to minimizethe transmission delay (sometimes the required infor-mation is not even accessible when it is needed dueto a disconnected wireless network). The idea ofprefetching is that a system delivers certain contentto a local mobile device that will likely be accessed bythe user soon. However, the prefetching decision iscontingent upon different network conditions andthe likelihood that the user will access this contentshortly. Prefetching operations consume the alreadylimited bandwidth of wireless networks and limitedstorage of mobile devices, and users often must payfor the service. It will be particularly expensive if theuser does not view the prefetched content eventually.Therefore, it is vital not only to determine what con-tent should be prefetched and when, but also tobuild utility models to analyze the trade-off betweenthe potential benefit of prefetching and cost undercurrent circumstances (for example, network condi-tions). For instance, [5] proposes a prefetching deci-sion scheme that utilizes the access probabilities andprefetch thresholds. The access probabilities indicatehow likely certain content will be requested by a user

Figure 1. Adaptivedisplay of Web

content with reconfigurations.

Figure 2. A DOM-treenavigation of the MSNhome page via Mobile

Web.

Figure 1(a) Figure 1(b) Figure 1(c) Figure 1(d)

Figure 1. Adaptive Display of Web Content with Reconfigurations Zhang fig 1 (2/07)

Figure 1. Adaptive Display of Web Content with Reconfiguration.

Figure 1 (a) Figure 1 (b) Figure 1 (c) Figure 1 (d)(a) (b) (c) (d)

encoded in an appropriate way so that it can bedecoded on the client side.

• Mobile devices have limited memory capacity,which may be insufficient for viewing multimediacontent, particularly video.

When a user accesses multimedia content on amobile device, a major cost is the long transmissionlatency due to the large data size, limited networkbandwidth, and low local memory. As a result, it isessential to develop effective adaptation approachesto minimizing this cost while maintaining a certainlevel of multimedia content quality. There are severalkey requirements for multimedia content adapta-tion: automated adaptation without user interven-tion; graceful quality degradation; seamless handoffsacross networks during roaming; and high QoS withlow jitter, delay, and guaranteed bandwidth. From aQoS perspective, sometimes users do not need abest-quality rendition, but rather good-enough qual-ity to convey the needed information [8]. MinimumQoS should be defined to support smoothaudio/video playback in a wireless environment. Aset of quality attributes (for example, color, imageresolution, frame rate, scaling, or modality) can bespecified for different types of multimedia contentfor QoS-based content adaptation.

Content providers can encode multimedia contentwith specific parameters, or a media converter canconvert it to another format to better fit the deliverycontext. For example, in a traffic-heavy network, anadaptive system may regenerate content with fewercolors and lower resolutions in order to reduce trans-mission latency. It can also switch media modality inorder to suit the current network bandwidth.

A variety of approaches can be applied to imple-ment adaptation for multimedia content [9]:

• Multiple encoding. Here, servers store multiplecopies of the same image or audio/video streamwith different encoding methods or parameters.Applications dynamically choose an appropriatecopy based on current network condition.

• Transcoding. This refers to the mapping of a non-scalable stream to another non-scalable streamwith a different compression rate (for example,MPEG to H.263). Transcoding avoids the neces-sity of storing multiple files corresponding to var-ious bit rates. Since this approach requiresdecoding and recoding of each media stream, it iscomputation-intensive and requires deep under-standing of different encoding and decodingschemes.

• Layered encoding (transcaling and multicasting).

This is probably the most popular way to provideadaptation for multimedia content. Thisapproach usually generates a base layer and one ormore enhancement layers to cover the desiredbandwidth range. Layers can be added ordropped by joining or leaving a multicast group.

• Rate shaping, or allowing an application to changemedia encoder parameters such as image resolu-tion, video frame rate, quantization parameter, andmovement detection threshold.

We believe that enabling access to adaptive, person-alized, and context-aware information and service iscrucial to fully take advantage of the pervasivenessand convenience of mobile devices. It will not onlyencourage users to fully embrace mobile technolo-gies, but also promote development of innovativemobile applications.

References1. Adipat, B. and Zhang, D. Adaptive and Personalized Interfaces for

Mobile Web. Proceedings of the 15th Annual Workshop on InformationTechnologies and Systems (WITS’05). Las Vegas, NV, 2005. 21–26.

2. Billsus, D., Pazzani, M. and Chen, J. A learning agent for wireless newsaccess. In Proceedings of the International Conference on Intelligent UserInterfaces. (New Orleans, LA, Jan. 9–12, 2000), 33–36.

3. Buyukkokten, O., Garcia-Molina, H., and Paepcke, A. Seeing thewhole in parts: Text summarization for Web browsing on handhelddevices. In Proceedings of the 10th International WWW Conference.Hong Kong (May 1–5, 2001).

4. Cheverst, K., Davies, N., Mitchell, K., and Friday, A. Experiences ofdeveloping and deploying a context-aware tourist guide: The GUIDEProject. In Proceedings of the 6th Annual International Conference onMobile Computing and Networking. (Boston, MA, Aug. 6–11, 2000),20–31.

5. Jiang, Z. and Kleinrock, L. Web prefetching in a mobile environment.IEEE Personal Communication (Oct. 1998), 25–34.

6. Laakko, T. and Hiltunen, T. Adapting Web content to mobile useragents. IEEE Internet Computing (March–April, 2005), 46–53.

7. Lank, E. and Phan, S. Focus+context sketching on a pocket PC. In Pro-ceedings of ACM CHI 2004 (Apr. 24–29, 2004, Vienna, Austria),1275–1278.

8. Lum, W. and Lau, F. User-centric content negotiation for effectiveadaptation service in mobile computing. IEEE Transactions on SoftwareEngineering 29, (2003), 1100–1111.

9. Mao, Z., So, H., Kang, B., and Katz, R. Network support for mobilemultimedia using a self-adaptive distributed proxy. In Proceedings of the11th International Workshop on Network and Operating Systems Supportfor Digital Audio and Video. Port Jefferson, NY, (June 25–26, 2001),107–116.

10. Yang, C.C. and Wang, F.L. Fractal summarization for mobile devicesto access large documents on the Web. In Proceedings of the Interna-tional WWW Conference (Budapest, Hungary, May 20–24, 2003),215–224.

11. Zhang, D. and Adipat, B. Challenges, methodologies, and issues in theusability testing of mobile applications. International Journal of HumanComputer Interaction 18, 3 (2005), 293–308.

Dongsong Zhang (zhangd@umbc.edu) is an assistant professor ofinformation systems at the University of Maryland in Baltimore.

© 2007 ACM 0001-0782/07/0200 $5.00

c

COMMUNICATIONS OF THE ACM February 2007/Vol. 50, No. 2 79

any adaptation. A user must use both vertical andhorizontal scroll bars in order to see the entire con-tent, which can be cumbersome and may cause loss ofcontext.

There are some simple ways to adapt content pre-sentation based on the features of a device and user’spreference to improve its navigation. For example, asshown in Figure 1(c), a “Fit to Screen” function canbe activated to produce a scaled-down version (Figure1(d)) of the original Web page to better fit the widthof a device through syntactic translation (withoutremoving any content from the original page); imagesembedded in a Web page and the Internet address barcan be removed to reducethe page size and increasedisplay space; and thefont size of textual con-tent can be adjusted bythe user. Such adaptationpreferences, oncedefined, can be stored inthe user profile and auto-matically used for futureadaptation.

Various summariza-tion and visualizationtechniques have also beenused in content presenta-tion adaptation formobile devices [7, 10]. Web pages often containmany sections and not all of them are of interest tousers. It is argued that the document summarizationon handheld devices should make use of “hierarchicaldisplay,” which is suitable for navigation of a largedocument and ideal for small area display [10]. AWeb page can be organized into a multi-level hierar-chy with a thumbnail representation at the upperlevel for providing a global view and an index to a setof sub-pages at the lower level for details [3]. A usercan select a desired portion of a Web page to zoom infor further details.

We have developed a Mobile Web system thatautomatically adapts the presentation of a Web pageon a mobile device based on a three-tier architecture[1]. In addition to adaptation functions that users canactivate, as shown in Figure 1(c), Mobile Web alsofeatures DOM (Document Object Model)-tree navi-gation, summarization, personalized topic spotting,and fisheye view, aiming to make Web browsing andinformation seeking on handheld devices more effec-tive. By following the principle of “overview first,then zoom in for details,” Mobile Web first automat-ically parses a Web page and generates a DOM-treehierarchical view of its content, as shown in Figure 2.

The main task of generating a DOM-tree is toidentify content blocks and their relationships in aWeb page and to extract labels that can representthose content blocks. A user can either expand anybranch of the tree, or view the summary of a selectedsection that is dynamically generated based on heuris-tic rules. Based on the summary, the user can deter-mine if he or she wants to see the full content of thatsection. Mobile Web stores users’ personal interest askeywords in their user profiles. When a Web page isdisplayed, those specified keywords appearing in thepage will be automatically highlighted with differentcolors for easy spotting, and the page display is auto-

matically anchored bythose keyword occurrences.Focus+context visualizationtechniques are designed tointegrate a high-detail focusarea and a low-detailperiphery in order to maxi-mize display space. Via acustomized fisheye viewcomponent in MobileWeb, users can navigatecontent through a focus-context view, with the focalcontent displayed in a

larger font size, while peripheral content is still shownin the surrounding area with reduced granularity ofdetail, such as a smaller font size (see Figure 3).

Environmental context can also be used to adaptthe content presentation. For example, when the lightlevel is low or the device is on the move (for example,the user is walking), then the font size of textual con-tent can be automatically enlarged to ease reading.

ADAPTATION OF MULTIMEDIA CONTENT

With advances in multimedia and wireless network-ing technologies, there is great potential to delivermultimedia content (for example, images, audio,and video) to mobile devices. NBC, for example, isconsidering delivering some TV programs to sub-scribers. Constraints of mobile devices and wirelessnetworks, however, pose several challenges to multi-media-based mobile entertainment:

• Low and varied network bandwidth will causelong latency while transmitting multimedia con-tent over wireless networks.

• A mobile device will not be able to view an audioor video stream if it does not have enough decod-ing power.

• Different mobile devices support different mediaformats. Video on a streaming server must be

78 February 2007/Vol. 50, No. 2 COMMUNICATIONS OF THE ACM

Figure 3. Analysis results.

Without fisheye With fisheye

encoded in an appropriate way so that it can bedecoded on the client side.

• Mobile devices have limited memory capacity,which may be insufficient for viewing multimediacontent, particularly video.

When a user accesses multimedia content on amobile device, a major cost is the long transmissionlatency due to the large data size, limited networkbandwidth, and low local memory. As a result, it isessential to develop effective adaptation approachesto minimizing this cost while maintaining a certainlevel of multimedia content quality. There are severalkey requirements for multimedia content adapta-tion: automated adaptation without user interven-tion; graceful quality degradation; seamless handoffsacross networks during roaming; and high QoS withlow jitter, delay, and guaranteed bandwidth. From aQoS perspective, sometimes users do not need abest-quality rendition, but rather good-enough qual-ity to convey the needed information [8]. MinimumQoS should be defined to support smoothaudio/video playback in a wireless environment. Aset of quality attributes (for example, color, imageresolution, frame rate, scaling, or modality) can bespecified for different types of multimedia contentfor QoS-based content adaptation.

Content providers can encode multimedia contentwith specific parameters, or a media converter canconvert it to another format to better fit the deliverycontext. For example, in a traffic-heavy network, anadaptive system may regenerate content with fewercolors and lower resolutions in order to reduce trans-mission latency. It can also switch media modality inorder to suit the current network bandwidth.

A variety of approaches can be applied to imple-ment adaptation for multimedia content [9]:

• Multiple encoding. Here, servers store multiplecopies of the same image or audio/video streamwith different encoding methods or parameters.Applications dynamically choose an appropriatecopy based on current network condition.

• Transcoding. This refers to the mapping of a non-scalable stream to another non-scalable streamwith a different compression rate (for example,MPEG to H.263). Transcoding avoids the neces-sity of storing multiple files corresponding to var-ious bit rates. Since this approach requiresdecoding and recoding of each media stream, it iscomputation-intensive and requires deep under-standing of different encoding and decodingschemes.

• Layered encoding (transcaling and multicasting).

This is probably the most popular way to provideadaptation for multimedia content. Thisapproach usually generates a base layer and one ormore enhancement layers to cover the desiredbandwidth range. Layers can be added ordropped by joining or leaving a multicast group.

• Rate shaping, or allowing an application to changemedia encoder parameters such as image resolu-tion, video frame rate, quantization parameter, andmovement detection threshold.

We believe that enabling access to adaptive, person-alized, and context-aware information and service iscrucial to fully take advantage of the pervasivenessand convenience of mobile devices. It will not onlyencourage users to fully embrace mobile technolo-gies, but also promote development of innovativemobile applications.

References1. Adipat, B. and Zhang, D. Adaptive and Personalized Interfaces for

Mobile Web. Proceedings of the 15th Annual Workshop on InformationTechnologies and Systems (WITS’05). Las Vegas, NV, 2005. 21–26.

2. Billsus, D., Pazzani, M. and Chen, J. A learning agent for wireless newsaccess. In Proceedings of the International Conference on Intelligent UserInterfaces. (New Orleans, LA, Jan. 9–12, 2000), 33–36.

3. Buyukkokten, O., Garcia-Molina, H., and Paepcke, A. Seeing thewhole in parts: Text summarization for Web browsing on handhelddevices. In Proceedings of the 10th International WWW Conference.Hong Kong (May 1–5, 2001).

4. Cheverst, K., Davies, N., Mitchell, K., and Friday, A. Experiences ofdeveloping and deploying a context-aware tourist guide: The GUIDEProject. In Proceedings of the 6th Annual International Conference onMobile Computing and Networking. (Boston, MA, Aug. 6–11, 2000),20–31.

5. Jiang, Z. and Kleinrock, L. Web prefetching in a mobile environment.IEEE Personal Communication (Oct. 1998), 25–34.

6. Laakko, T. and Hiltunen, T. Adapting Web content to mobile useragents. IEEE Internet Computing (March–April, 2005), 46–53.

7. Lank, E. and Phan, S. Focus+context sketching on a pocket PC. In Pro-ceedings of ACM CHI 2004 (Apr. 24–29, 2004, Vienna, Austria),1275–1278.

8. Lum, W. and Lau, F. User-centric content negotiation for effectiveadaptation service in mobile computing. IEEE Transactions on SoftwareEngineering 29, (2003), 1100–1111.

9. Mao, Z., So, H., Kang, B., and Katz, R. Network support for mobilemultimedia using a self-adaptive distributed proxy. In Proceedings of the11th International Workshop on Network and Operating Systems Supportfor Digital Audio and Video. Port Jefferson, NY, (June 25–26, 2001),107–116.

10. Yang, C.C. and Wang, F.L. Fractal summarization for mobile devicesto access large documents on the Web. In Proceedings of the Interna-tional WWW Conference (Budapest, Hungary, May 20–24, 2003),215–224.

11. Zhang, D. and Adipat, B. Challenges, methodologies, and issues in theusability testing of mobile applications. International Journal of HumanComputer Interaction 18, 3 (2005), 293–308.

Dongsong Zhang (zhangd@umbc.edu) is an assistant professor ofinformation systems at the University of Maryland in Baltimore.

© 2007 ACM 0001-0782/07/0200 $5.00

c

COMMUNICATIONS OF THE ACM February 2007/Vol. 50, No. 2 79

any adaptation. A user must use both vertical andhorizontal scroll bars in order to see the entire con-tent, which can be cumbersome and may cause loss ofcontext.

There are some simple ways to adapt content pre-sentation based on the features of a device and user’spreference to improve its navigation. For example, asshown in Figure 1(c), a “Fit to Screen” function canbe activated to produce a scaled-down version (Figure1(d)) of the original Web page to better fit the widthof a device through syntactic translation (withoutremoving any content from the original page); imagesembedded in a Web page and the Internet address barcan be removed to reducethe page size and increasedisplay space; and thefont size of textual con-tent can be adjusted bythe user. Such adaptationpreferences, oncedefined, can be stored inthe user profile and auto-matically used for futureadaptation.

Various summariza-tion and visualizationtechniques have also beenused in content presenta-tion adaptation formobile devices [7, 10]. Web pages often containmany sections and not all of them are of interest tousers. It is argued that the document summarizationon handheld devices should make use of “hierarchicaldisplay,” which is suitable for navigation of a largedocument and ideal for small area display [10]. AWeb page can be organized into a multi-level hierar-chy with a thumbnail representation at the upperlevel for providing a global view and an index to a setof sub-pages at the lower level for details [3]. A usercan select a desired portion of a Web page to zoom infor further details.

We have developed a Mobile Web system thatautomatically adapts the presentation of a Web pageon a mobile device based on a three-tier architecture[1]. In addition to adaptation functions that users canactivate, as shown in Figure 1(c), Mobile Web alsofeatures DOM (Document Object Model)-tree navi-gation, summarization, personalized topic spotting,and fisheye view, aiming to make Web browsing andinformation seeking on handheld devices more effec-tive. By following the principle of “overview first,then zoom in for details,” Mobile Web first automat-ically parses a Web page and generates a DOM-treehierarchical view of its content, as shown in Figure 2.

The main task of generating a DOM-tree is toidentify content blocks and their relationships in aWeb page and to extract labels that can representthose content blocks. A user can either expand anybranch of the tree, or view the summary of a selectedsection that is dynamically generated based on heuris-tic rules. Based on the summary, the user can deter-mine if he or she wants to see the full content of thatsection. Mobile Web stores users’ personal interest askeywords in their user profiles. When a Web page isdisplayed, those specified keywords appearing in thepage will be automatically highlighted with differentcolors for easy spotting, and the page display is auto-

matically anchored bythose keyword occurrences.Focus+context visualizationtechniques are designed tointegrate a high-detail focusarea and a low-detailperiphery in order to maxi-mize display space. Via acustomized fisheye viewcomponent in MobileWeb, users can navigatecontent through a focus-context view, with the focalcontent displayed in a

larger font size, while peripheral content is still shownin the surrounding area with reduced granularity ofdetail, such as a smaller font size (see Figure 3).

Environmental context can also be used to adaptthe content presentation. For example, when the lightlevel is low or the device is on the move (for example,the user is walking), then the font size of textual con-tent can be automatically enlarged to ease reading.

ADAPTATION OF MULTIMEDIA CONTENT

With advances in multimedia and wireless network-ing technologies, there is great potential to delivermultimedia content (for example, images, audio,and video) to mobile devices. NBC, for example, isconsidering delivering some TV programs to sub-scribers. Constraints of mobile devices and wirelessnetworks, however, pose several challenges to multi-media-based mobile entertainment:

• Low and varied network bandwidth will causelong latency while transmitting multimedia con-tent over wireless networks.

• A mobile device will not be able to view an audioor video stream if it does not have enough decod-ing power.

• Different mobile devices support different mediaformats. Video on a streaming server must be

78 February 2007/Vol. 50, No. 2 COMMUNICATIONS OF THE ACM

Figure 3. Analysis results.

Without fisheye With fisheye