Visualization

Jim Foley & Chris Connolly

Information visualization, like its cousin scientific data visualization, is a powerful way to effectively convey understanding to the user. Recognizing that the WWW is an enormous graph of nodes and arcs, we imagine various overview and navigational views of WWW sub-structures being presented to the user to assist in exploratory resource discovery. For instance, a graph showing the relationship of a home page to all of its subsidiary pages would provide a user with a valuable "road map" to all of the pages. However, the paucity of explicit content semantics (meta-information about nodes and links) in HTML makes creating meaningful overview maps quite difficult.

Recommendation 1: Research on Information Road Maps

Support research on ways to visualize WWW-like information spaces. To do this, at least three issues need to be addressed: 1) What types of "road maps" are useful for navigating? 2) What descriptive semantics must be available in HTML or related structures (link databases, structures above HTML, etc.) to allow the road maps to contain meaningful semantics? 3) How can the road maps be automatically generated from the semantics?

A reasonable place to start this work is with graph-like diagrams, but graphs are not the only visual metaphor for information space navigation. The way-finding literature in architecture, map-making, book design, and information design all offer rich sets of concepts. Now is the time to explore such metaphors to understand their applicability. Those which show promise should be tried, both in formal experiments and experientially. Means to automatically generate an appropriate metaphor for a particular information space will be needed: such generation includes both creation of the visual representation itself as well as the automatic binding of individual information elements to the visual metaphor.

Because information spaces are so large, an important element of creating overviews will be creating appropriate visual abstractions which convey meaningful relationships while eliding unnecessary detail. This in turn will often require semantic descriptors beyond what is found in HTML or in other meta languages.

Recommendation 2: Visualization of Bibliographic Search Results

Support research to develop and experiment with a wide variety of ways to present and manipulate search results. Presenting the user with a linearly-ordered list of documents based on some notion of "relevance" is quite abstract and does not help the user understand the high-dimensional space within which relevance is defined. Various scatter plot and other types of presentations have been developed [Heath 95; Rao 95; Verasamy 95]; exploratory research is needed to develop others, and experimental studies are sorely needed to understand the strengths and weaknesses of the existing visualizations. A set of benchmark tasks against which to measure various visualizations is another needed research outcome.

Most bibliographic search facilities currently used on the Web rely on string search rather than more sophisticated database organization and query techniques. Integration of more advanced information retrieval tools into the Web would enhance bibliography management and sharing. The set of current visualization tools for retrievals should be integrated into WWW search engines, as a way of exposing a wider community to the state of the art and hence engaging their imaginations in creating new tools.

Recommendation 3: Web Viewers as the New Desktop Metaphor

Another aspect of visualization deals with the potential for expanding the role of Web viewers. Web viewers present to a computer user a collection of resources which are available on other people's computers - currently, on other Web servers. At the same time, the user has to look, in other windows, at a different collection of different visual representations of other types of resources: programs, files, directories, email messages, etc. located on his or her own computer. This asymmetry is paradoxical. After all, one of the key Web concepts is the URL, the Universal Resource Locator. Why should URLs not be used to locate resources in a truly universal manner, including those on one's own workstation? Why should a Web viewer not present the total set of resources available to a user? Hyper-G provides a local file browser [Kappe 93].

The idea, then is to use the Web and its viewers as a single, unified mechanism for presenting and accessing information, eliminating the artificial separation between one's own local files which are not integrated into a Web server, and universal resources, which are integrated.

This has several implications, some of which deal with visualization and presentation, others of which go deeper within the Web infrastructure. At the presentation level, it reinforces the need for a research agenda dealing with visual metaphors for presenting, understanding, organizing, and navigating rich information spaces. Note that visual interfaces to computer operating systems, such as the Macintosh OS and Microsoft Windows, provide simple but inadequate such metaphors. There is no reason to think that other metaphors are not available. Global file systems, like Prospero [Neuman 92], which provide simplistic browsing, are indeed research in the right direction.

Within the Web infrastructure, the "Visualization as universal resource locator" concept means that programs have to become first-class Web objects, objects which can be searched for and loaded and executed, just as text files can be searched for and loaded. In this way the Web becomes a browsable repository for programs, opening up local file systems to authorized users via a single, powerful mechanism. Incorporation of programs as objects would be a useful adjunct to the active paper concept, since each program is a complete algorithm specification. This greatly facilitates the reproduction and sharing of research results in applied disciplines such as robotics and computer vision.

Recommendation 4: Scaling the Visualization to Different Platforms

Access to the Web is currently from standard computing platforms, e.g. UNIX workstations, PCs, Macs, etc. With the increasing use of PDAs, PC-based TV set-top boxes, and mobile computers communicating via limited bandwidth, research is needed to develop methods of automatically scaling, or adapting, Web information to these new platforms. Also, as more and more individuals with handicaps access information via computers, the same scaling concepts can be used to adapt presentations to assist those with limited sight or limited motor skills. We discuss here three aspects to scalability: display real-estate scaling, bandwidth scaling, and auditory scaling.

Real estate scaling is concerned with the limited screen resolution of PDAs and home TVs. The research agenda here includes developing rule-based information presentation tools whose task is to automatically render information in ways which make sense for the targeted device - by eliding detail, by creating hierarchical levels of selection where none previously existed, by spatially rearranging information, by whatever it takes to allow the user to sensibly access and see (or hear) information.

Bandwidth scaling refers to being able to deal with limited bandwidth between the information source (server) and information destination (client). This is an issue even for workstations connected at gigabit or ethernet speeds, and becomes even more of a concern for a 19.2Kbps wireless connection or a 9.6Kbps home telephone connection (we believe the demise of this low-speed access will be much slower than predicted). One approach to bandwidth scaling is multiresolution retrieval and display, whereby information displayed to the user is incrementally refined as more data is retrieved and components are received at progressively higher resolutions. Each information object published by a server may be defined at various resolutions according to its data type. For a digitized photograph, a low resolution version may have several adjacent pixels grouped together and assigned an average color value (e.g., an image pyramid). For a large Postscript document, a low resolution copy may consist of the abstract and the index. Depending on the typical load on the server and the storage requirements, an agent at the server site can decide to store one or more of the multiresolution versions and generate the others on demand. In the default mode of object transfer, a low resolution copy of the object is retrieved and displayed to the user while the full or higher resolution copy is being transferred. The requesting user is thus shown a version of the object promptly and can decide whether to continue with the retrieval or to abort the request. This can noticeably reduce the `dead time' spent by users as they wait for network connections to be made and the requested information to be transferred. Similar retrieval and caching schemes are already used in some image analysis systems, but can be generalized in the Web to apply to many different object types.

Audio scaling is the concept of either replacing or supplementing visual information with auditory cues (or vice versa), or adapting audio to monaural, stereophonic, quadraphonic outputs and to different acoustic environments.

In the coming days of ubiquitous and mobile computing, as an individual frequently moves from use of one platform to another, all types of scaling used will need to be consistent and provide seamless integration across platforms.

Tools to automate the process of building views will help make the Web more scalable with respect to the rate of generation of new data and bandwidth. Agents may be deployed at the user's platform to scan the information that has newly become available from a set of given sources and selectively incorporate it into the user's personal view according to a specified criterion.

The navigational views of information space will need to be integrated with this notion of multiresolution retrieval as well. In a very real sense, a visual map of an information space is simply a low resolution representation of the space. Intelligent agents will be needed to create all sorts of abstracted views of individual information documents and of documents organized in an information space.

Recommendation 5: Tools to Develop Context-Sensitive Visualizations

Database systems have long applied the concept of subschemas to provide multiple views of the same underlying information. One way to think of multiple views is in the context of the scalability issues just discussed. But another way to think of multiple views is in the context of access-path view dependency. For example, the GVU Center at Georgia Tech has an extensive set of pages with a distinctive GVU look <URL:http://www.cc.gatech.edu/gvu/>. However, the home departments of GVU faculty and students all have home pages with their own looks. It is unreasonable to ask everyone to have multiple sets of pages, one with the GVU look and one for their home department's look, with the pages having duplicate information but having different graphic design looks and different logical organizations of information onto pages and different linkage structures. What is needed is a way to separate content from appearance, organization, and linkages so that the access path to an information collection can be used to determine which "view" to apply to the underlying information. Note that this notion of views is also relevant to the discussion of automatic reorganization in Section 5.7.

References

Heath L., Hix D., Nowell L., Wake W., Averboch G., & Fox E. (1995) Envision: A User-Centered Database from the Computer Science Literature. Communications of the ACM, 38(4), (in press).

Kappe F., Maurer H., & Scherbakov N. (1993) Hyper-G: A Universal Hypermedia System. Journal of Educational Multimedia and Hypermedia, 2(1) pp.39-66.

Neuman B. C. (1992) Prospero: A Tool for Organizing Internet Resources. Electronic Networking: Research, Applications and Policy, 2(1), pp.30-37.

Rao R., Pedersen J.O., Mackinlay J.D., Masinter L., Hearst M., Halvorsen P.-K., & Card S.K. (1995) Towards Rich Interaction in the Digital Library. Communications of the ACM, 38(4), (in press).

Veerasamy A., Hudson S., & Navathe S. (1995) Visual Interface for Textual Information Retrieval System. IFIP 2.6 Working Conference on Visual Database Systems - 3.

Table of Contents

Next Chapter