Adaptive Interfaces Literature Review

Adaptive Interfaces Literature Review

Ryan Getek

Spring 2007

Index

1Introduction

1.1Usability

1.2Why News and Entertainment Sites

2Adaptation

2.1User vs. Interface

2.2Profiling

2.3How Interface Data Can Be Obtained

2.4Where adaptation occurs

2.5When to adapt

2.6How content can be adapted

2.6.1Screen Adaptation

2.6.2Network Connection Speed Adaptation

2.6.3Browser Type Adaptation

3Adaptation Vs. Personalization

3.1Security and Privacy

3.2The Problems with Personalization

4Conclusion

5References

1Introduction

News and entertainment websites such as MSN.com, CNN.com, and news.Google.com present a significant amount of information to the user in a single page. Yet, few take advantage of rapidly increasing screen sizes that are becoming available as LCD monitor technology advances and prices drop for large displays. In most cases, the page remains a fixed size width of information that is either fixed in the left portion or the middle of the screen. In addition, increased demands for multimedia content by users have encouraged site developers to create content with larger file sizes that may be difficult to download when the user has a slow Internet connection. Varying browser types further complicate development for content in these types of sites because a page may be displayed differently to users with different browsers. Adaptive interfaces offer a potential solution for these issues and a way to enhance usability by modifying the presentation of information based upon an understanding of the interface.

There is a broad range of interpretation for the term ‘adaptive interfaces’. Adaptive approaches have beenproposedbothfor desktop applications and web interfaces.[1, 2] Web interface adaptation becomes potentially even more complex as it is often combined with personalization. The scope is refined within the context of this review to adaptation of presentation within the web interface rather than traditional interpretations of personalization which include modifying the content and focusing on the specific user. We define an adaptive interface as “the automatic modification of the presentation of information to users based upon the physical characteristics of the user’sscreen, network connection speed, or browser type”. While the potential application of adaptive techniques to network characteristics and browser type will also be explored, the screen component will form the core of this review. The definition implicitly distinguishes between user characteristics and interface characteristics, though user preferences are also within the scope of this review.

One of the reasons that such a distinction is necessary is that the increased awareness and importance of maintaining security and privacy on the Internet has begun to affect how web content providers can collect, store, and share user information. The difficulties associated with obtaining information about the user and implementing a personalized interface has proven difficult for a variety of reasons in the past. While traditional Human Computer Interaction (HCI) knowledge has focused on enhancing usability by seeking to understand the individual user, it may become necessary to generalize an understanding of user characteristics when factoring in the security and privacy concerns as well as problems encountered when trying to personalize content to user interests and characteristics. These issues will be addressed in depth later.

The collective impact of the security, privacy, and personalization issues is contributing to a paradigm shift from traditionally accepted usability practice where knowledge of the user was most important. By collecting information about the interface instead and adapting presentation based upon generally applicable models of what presentation characteristics maximize usability, potentially invasive collection of data about individual users can be avoided. This approach can maximizes usability from a utilitarian perspective as well, because instead of providing high levels of personalization to a smaller group of users who choose to take advantage of such functionality or whose habits facilitate the collection of relevant data, the adaptive approach based upon interface characteristics enables usability benefits to be provided across a much broader range of users.

Two important questions must be answered in order to determine whether web content developers should adopt adaptive techniques. The first question is whether concerns over security and privacy as well as problems with existing personalization approaches are significant enough to warrant shifting the focus of data gathering from the user to the interface. The second is whether adapting the interface actually increases usability. As part of answering the second question, we will define usability within the news and entertainment website context.

1.1Usability

In order to measure usability and describe the impacts resulting from the implementation of adaptive interfaces, the term usability must first be defined and metrics must be identified for how it can be measured. A basic set of measurable attributes will be defined so that the context of usability as it is used within this review can be better understood.

The most commonly used measures of usability are task accuracy, task speed, and subjective measures of user satisfaction. These measures provide an important balance because a user might like a particular interface better and give it higher subjective ratings, but perform poorly in tasks while using this interface. Whether the task completion or subjective evaluation is most important depends upon the user’s purpose for completing the action.

With news and entertainment sites, the subjective satisfaction with the site is more important than it might be in other types of sites. For example, a healthcare site’s main function might be to provide the user with data about providers, specialties, coverage, and contact information. The ability to complete the task of finding that information is likely to be more important than the subjective ratings. The difference with news and entertainment web sites is that users have a different searching pattern because they may not even know what they want in many cases.

1.2Why News and Entertainment Sites

Users generally regard news sites among the most public of the types of sites they visit and are less concerned about other people finding out that they have visited such sites compared to health, banking, email, or most other types of sites.[3] The tendency of users to consider these site types less private than other types of site makes them an attractive target for adaptive techniques. In addition to the openness of news and entertainment sites, the following attributes make them conducive to adaptive approaches:

• The modularity of information is conducive to applying techniques such as the use of XML style sheets or Cascading Style Sheets (CSS) for one-time design, continuous reuse
• Large number of users and broad impact
• One of most commonly visited site types
• Lots of information is presented in the main page, versus other types of sites that might segment content more into separate pages

Many users set their default page as a news site, and browser plug-ins and instant messenger programs might even change the default page to msn.com, for instance, as part of the installation process. A large target audience and a significant amount of information displayed to users means that if the potential benefits of adaptive interfaces can be applied to these types of sites, the resulting impact will be extensive.

2Adaptation

The interpretation for the term adaptive interfaces has varied over time. In addition to these variations, there are also a wide range of applications and target audiences for adaptive interfaces. Hancock and Chignell, for instance, sought to match mental workload and task complexity in order to maximize the efficiency of the user in completing complex tasks.[4]

The importance and potential benefits have also been identified for enabling users to access data as they transition sessions across different types of devices, such as continuing to access a news story on a small laptop that was originally loaded on a desktop PC with a large screen.[5] The automatic adaptation can complement other session-aware efforts and support more efficiently transitioning between different device types while reducing the impact to the user. This research area becomes increasingly important as users start to adopt a wide range of device types and expect to be able to use them in a relatively uniform manner.

Adaptive interface development has historically faced some significant challenges. Some of the challenges noted by Norcio and Stanley in the 80’s included user confusion created by a changing interface, users who might intentionally provide inaccurate data, and complex design requirements with potentially high cost.[6] Currently, a more conducive environment exists for adaptive design becauseof the focus on interface characteristics rather than user characteristics, the enhanced technology behind enabling such systems, and the evolving web design practices that increasingly support adaptive techniques.

2.1User vs. Interface

Adaptive interfaces have been proposed as a potential usability enhancement for applications ranging from educational environments to airborne crew workstations.[7, 8] In many cases, the basis for adaptation varies widely and has traditionally ranged frominterfacecharacteristics, environment data, and context data all the way to extensive and often invasive user data. While the focus of this review is upon adapting to aspects of the interface rather than the characteristics of the user, significant work has been performed in the user area that provides important perspective on how the adaptation issues have historically been viewed and researched.

Physiological data has been used as an input to adapt an interface based upon the user’s response to the content presented and associated mental workload, including measures such as Galvanic Skin reflex, heart rate, and other indicators of physical state.[9, 10] Implementation of his work, such as in [11], built upon the framework developed by Norcio. Systemsthat measure physiological characteristics require some knowledge about the user’s level of expertise and even health condition in order to accurately differentiate between relevant effects and those created by external effects.[12] While the complexity, cost, and difficulty in interpreting human physiological responses makes these types of adaptive interfaces fairly difficult to build and maintain, the efforts in this area demonstrate the extent to which adaptive techniques can be incorporated into the system.

A fair amount of adaptive interface research has been performed for mobile devices and PDA’s, but significantly less work has been done in trying to apply the concepts to a broad range of PC’s.[13, 14] Much of the mobile work might be considered a reaction to the inadequacy of personalization and customization techniques for organizing content in a usable way on these devices, as a survey by Billsus et al. of wireless carriers in 2002 revealed that only 2-5% of wireless users leveraged such features.[15] The design considerations for the broader PC user base differ from the mobile concerns, though some mobile techniques such as Usage awaRe Interactive Content Adaptation (URICA) for mobile devices do adapt based upon the same interface attributes that are proposed here for PC’s.[16]

Srivastava et al. identify four distinct classes of web data including content, structure, usage, and user profiling.[17] While personalization and associated data mining techniques typically modify content with consideration for multiple of those attributes, adaptive interfaces in the context of this review focus on the structure and to a lesser extent, user profiling. The structure defines the presentation, and user profiling is based upon a general understanding of how the interface characteristics affect usability. In other words, while a profile is not generated for individual users, general profiles can be created for types of users such as users with a large monitor, fast Internet connection, and who use Internet Explorer 6. The issues related to profiling will be address further later in this review.

The greater the level of user knowledge that is obtained and implemented, the more tailored and accurate the implementation has to be. Inaccuracy can reduce usability compared to even non-personlized systems. For instance, if a system displays hockey content to a single-minded football fan, the personalization mechanism will likely reduce user satisfaction and therefore reduce usability. Such scenarios create a tradeoff between the expressiveness of the adaptation and the ease of use.[18] In addition, they also contribute to considerations in development. A system that is capable of modifying the content per user based upon such interests is likely more difficult to design and maintain than a system that can generalize to a broader segment of users. Furthermore, such narrow focus can alienate users when preferences are identified incorrectly or change over time.

Accessibility is an increasingly important consideration in web design. For government or other public websites, laws and standards of practice make facilitating access a requirement rather than a luxury. The Americans with Disabilities Act (ADA) establishes national guidelines that can be strengthened and more broadly applied through state laws. One example is how California’s extension of the ADA enabled a recent lawsuit to be filed. A University of California at Berkley student sued Target for failing to provide an accessible website for the visually impaired in a lawsuit that followed similar proceedings by other plaintiffs against America Online, Priceline, and Ramada based upon accessibility issues with their sites.[19]

The most important information the website needs to facilitate the best usability possible is the availability of assistive devices and user preferences for how to deliver the content to the devices and interface.[20] For instance, it is not as helpful to know that a user is visually impaired as it is to know that the user has a screen reader and prefers that content be represented as text rather than images. This is likely to differ between users based upon personal preference and level of disability. The site can design an adaptive version of the content better by knowing which interface and preference characteristics will most enhance the user’s ability to find and retrieve content.

Accessibility can be enhanced for users by enabling content to be displayed in a more usable manner without requiring special hardware. Dodd makes the case for enhancing intrinsic accessibility through adaptive techniques by making the pages better without requiring the user to do anything special or acquire special equipment.[21] This could be facilitated by more effectively using the space that is afforded by larger screens that users may already have.

Currently, a visually impaired user who buys a 24” widescreen monitor often still looks at a small ribbon of information in the middle of the screen. The user could adjust the browser’s text size attribute to make the text larger, but often the page becomes disproportioned and more difficult to interpret. Even worse, some site developers set a fixed text size in order to maintain the visual appeal of the design and avoid disorganization caused by varying browser text sizes. A study by Bartell found that although 82% of sites analyzed during the study facilitated viewing the text at 12 to 14pt, the recommended font size for maintaining legibility in continuous text, that knowing what size users were actually seeing the text was difficult or impossible.[22] Adaptive interfaces provide a mechanism by which accessibility could be enhanced without requiring any special equipment such as a screen reader for users with moderate levels of disability by using the screen area more effectively.

2.2Profiling

Significant differences exist between the need for profiling users for traditional personalization applications and for profiling users based upon an understanding of how they interact with different combinations of interfaces and presentation techniques for adaptive interfaces. To illustrate the difference, the profiling that is involved with personalization will be briefly discussed, then techniques for profiling in adaptive interface development will be analyzed.

Some systems seek to understand the user by recording and analyzing their actions and behaviors, such as the one discussed in [23]. These systems only provide benefits if the whole personalization chain remains intact. The data must be obtained by the service provider, it must be properly interpreted by the personalization engine, appropriate personalization techniques must be applied to the content to be delivered to the user, and ideally the user should be able to provide feedback. The breakdown in any part of this chain can lead to improper personalization with negative impacts to usability.

Web mining data such as log entries has been used for a variety of purposes from user pattern analysis to detecting problems with connectivity.[24] It can also be used to elicit user preferences or habits in order to tailor content to the user. The problem with this technique, as in many techniques that do not provide a transparent mechanism for collecting and applying user data to the resulting interface, is that a one time action may cause the system to deliver a mismatched set of content to the user.

Once a method for collecting data has been obtained, the system must process it in a meaningful way. One technique for applying the data to an personalization implementation is through the use of fuzzy logic, such as in [25]. Fuzzy logic techniques attempt to allow the system to learn from user actions and fluidly deliver tailored content based upon the habits of the user.