1990 levels) if nothing is done to reduce them. Access to equipment.

The Distributed Laboratory
Project at San Diego Supercomputing Center (SDSC) provides remote access to the intermediate voltage electron microscope at the University of California, San Diego for real-time data acquisition, visualization, and data analysis. Scientistswithout local access to this sophisticated and expensive equipment can thus pursue investigationsotherwise closed to them. Airplane design. Boeing’s 777 airplane was designed electronicallywithout physical drawings and models, using visualization. Moreover, designers at disparate locations reviewed and discussed model structures using recently developed visualization and network collaboration capabilities.

widely. We need to learn how to create flexible user interfaces, navigation tools, and search methods appropriate to the existing types of users and applications. Visual computing and display can help in symbolic representation of objects to manipulate. They also support flexible interfaces that can be adjusted to suit users with different preferences and cultural or educational backgrounds. These capabilities will help make the GI1 truly global.

References 1. N. Gershonand S.G. Eick, “Visualization’s New Tack: Making Sense of Information,” IEEESpectrum, Nov. 1995,pp.
38-56. 2. J. Brown et al., Visualization: Using Computer Graphics to ExpIore Data and Present Information, John Wiley & Sons, New York, 1995.

Empowering a responsible GI1
The power of the GI1 also carries a responsibilityto see that financially,technically, or physically disadvantaged persons are not left behind. Users and applications vary

Retrieved Information: A Survey
learly, the presentation method for information retrieved from the global information infrastructure (GII) makes a big difference to users. The NCSA Mosaic interface, for example, with its point-and-click multimedia page presentation, swelled popular interest in the World Wide Web. This experience suggests the possibility of increased usefulness if we apply visualization techniques to information retrieved from the GII. Note that, although the term “information visualization” is coming into use, the goal is really “information perceptualization.” The latter implies a richer use of many senses, including sound and touch, to increase the rate at which people can assimilate and understand information. In discussing the visualization of retrieved information, it helps to consider four functional levels: (1)the infosphere, (2) the workspace, ( 3 ) sensemaking tools, and (4) the document. This simple classificationlets us separate the functions served by the visualizations from the techniques themselves, which can be applied across functional levels.

Stuart K. Card Xerox PARC

C

lnfosphere
The infosphere (variously called the docuverse or docuspace, or sometimesjust cyberspace) is the reachable space of information sources,largely databases and documents. Of course, a visualization of the infosphere does not presuppose that all elements in the space are visible at one time (especially given a space the size of the GII) .Logically, the infosphere includes repositories, places where collections of information reside. Whether this structure is represented in thevisualization is a current research issue. How might we visualize the infosphere? The galaxy visualization method of Battelle Pacific Northwest National Laboratory’ uses multidimensional scaling of the similarities between documents (based on the correlation between their vector representations) to plot one point in Euclidean space for each document. These points form a semantic scatterplot. Similar documents tend to lie near each other, although the axes ofthe space aren’t always easy to interpret. The galaxy can also be projected onto a plane and elevation used to code frequency of mention, forminga semanticlandscape of doc-

IEEE Computer Graphics and Applications

63

Ironically. users can either occupy a space or get an overview of all the space. Feiner’s “worlds within worlds” method3for multidimensional databases falls somewhere between these two. instead of 2 using spatial coordinates. creates a semantic space of news stories whose visual structure is not fixed. Words previously rejected may appear in the context of words now selected. The Galaxy of News system’ 64 March 1996 . which provides a dynamic 3D workspace. but changes as the user moves.6Even more radically. a sixdimensional object. Zooming in on a particular word causes other words. and yet more words appear. a system with a similar name uses an almost opposite approach. In contrast to the representation of documents as points or clouds in a multidimensional space. to appear. say. parts of it appear to enlarge and shrink according to distance from the user. The user sees an arrangement of words derived from the news stories.^ Or. that is. as in the Narcissus system. Although the user experiences a space. there is no stable spacewhat is stable is the path. x Works pace A workspace is a collection of objects whose access is arranged to make relevant tasks efficient. Lamping and Rao7recast a linked network as a tree. 2 With 3D rooms. docux e ments can be visualized as linked 2 graphs or trees in two or three dimensions. Benedikt4 proposed representing the entire infosphere in a similar manner. Kohonen maps divide the available space into semantic region^. then plot the tree in hyperbolic space. by increasing the information-carryingabilities of periph- uments (see Figure 2 in Gershon’ssection ofthis report). The design of a workspace induces a cost structure over it. In his technique. a set of times associated with accessing and manipulating its objects Visualization methods can improve this cost strucl ure by enabling more information in a display.Special Report 1 The Web Forager. representing finer classifications of the story. As the user traverses the landscape. taking on new. The user zooms towards the relevant word. Workspaces allow multiple objects to be compared and grouped or combined. nuan ced m eanings. the nodes and links can be projected onto a non-Euclidean landscape. Other methods tackle how to represent parts of the infosphere. is plotted by having a new coordinate system for the last three dimensions sit inside the coordinate system for the first three. 3 The Film Finder.

these visual Figure 3 ) . Sensemaking often benefits from visualization techA second type of visualization. The user Spence and Apperly’s bifocal lens. or they based on Furnas’s work’’ expands only those parts of allow the construction of new information patterns from the tree related to some target node. Those movies that do meet the criteria the center and the periphery of the workspace. In large scale. Yet another technique provides a different scale for the slider moves.000 nodes representing part of the WWW and (b) the fish-eye version of the same cone tree.000 lines of code.Xerox WWW. Finally. The interest for actor.use coloring.”” This method.” the sides of the can click on dots representing movies to get details. mx Visualization techniques help present objects in a way that provides information across large zoom scales. or by altering visual proper. 5 A workspace can also be based on zooming.4 (a) Cone tree with 20. and less relevant parts of the tree shrink. director. by distortion. The workspace divides into “rooms” corresponding to user information working sets. I ” Large amounts of information can be placed k in the workspace by making them very small. Figure 2 shows another technique for building information-intensive workspaces. allows the user to cone tree of 20. Figure 1shows one version of a dynamic 3D workspace. the Web Forager. They re-represent est. an algorithm retrieved information to make patterns visible. defined points of interest. Colors represent how 5 Seesoft display of program code from AT&T Bell Labs. workspace are horizontally compressed to hold more Examples of dynamic movement and distortion include the cone tree and the perspective wall. position. tree that articulates to bring forward into focus those position nodes the user touches. The perinformation. Movies SeeSoft system16(see Figure 5) allows a user to view not meeting these specifications are visually culled as more than 12. When the users touches tion a moderate number of documents (50 to 120) some node in the tree. One type of visualization is acleverstaticview. the parts of the tree rotate. E IEEE Computer Graphics and Applications 65 . elerelative to a small number of spatially-positioned. and a gesture interface permits rapid interaction.ing the fish-eye techniques. a dynamic view.000 nodes representing the internal see the patterns of hits for many terms. The user moves sliders to indicate values of attributes form patterns quicklydetected by the eye. Figure 4b shows the same tree after applyple. occupy a space or get an overview of all thc spacc. The cone tree is a 3D (the Virtual Reality Modeling Language).* The 3D perspective allows more objects in the space. such as Spoerri’s “information crystal. The cone tree can be applied to a portion of the World old. by niques that let the user both span large-scalefactors and movement.”uch as projects or clients. as in Pad + + .but it uses spaces on more literal geometric spaces through VRML perspective instead of direct scaling to reduce the size of elements outside the focal area. Another exam.ments around the touched element grow. and by shifting load from slower cognitive to faster perceptual processes. era1parts of the display. Figure 4a shows a retrieval based on Boolean queries. Information can be f in more than one room at a time. an Wide Web infosphere by recasting the links as a tree elaborate form of Venn diagram for information according to a traversal algorithm. For example.spectivewall is a variant of the bifocal display. In the fish-eye version of the cone tree. In are plotted in a scattergraph of rating by year. changes in response to user action. and length of a movie. or other visual attributes to preties. The tree can be culled Sensemaking tools Sensemaking tools help users understand informa. considerable research involves basing work. and the user can either .manually or automatically to show only items of intertion by associating and combining it. the VIBE system. the Film Finder14employs culling (see sent classifylnginformation.13 uses word frequencies to posi. either by culling.term.

” s e rInteiface Software. The user inserts query terms.6 WebBook being fanned so that the user can rapidly examine a group of related WebPages.. 1994.Symp. 119-224. 95. Another version. Elenedikt. IEEE Computer Society Press. “AFocus + Context Technique Based on Hyperbolic Geometry for VisualizingLarge Proc. and G. 7. even fan them for rapid scanning (as in Figure 6).A.. Los Alamitos.J. such as a spreadsheet. “Cyberspace:Some Proposals. ous step is to create entities at a higher level than the page.. detailed focal information can be integrated with a scaled display.I8for example. they could be given I I u visualizationsensemakingtools for analyzingand invesi tigating the information. Gershon et a1. Mass. Superbook. 2.. 3-12. ACM Press. recently the code was modified. MITPress.e. pp. 6. Beshers. Rather than having separate focus and context displays. 1995. but could also represent programmer or department. consider visualization of individual documents.. Finally.Proc.”Proc.LAN takes ccntcr atage in ng. decorated in some way to show content. Science. Liebscher. count the eo-occurrences of words in a set of documents. “Visualizingthe Nonvisual: Spatial Analysis and Interaction with Information from Text Documents.47%.” in CHI 95. ACM Conf. 1991. Visiuiilization and user interfaces The research summarized here suggests how to improve the human interface to the Web. “Graphic Representation of Electronic Search Patterns. in.J a n a h “ A r m y tests prototypc baztlofield inforeai “Par=gon p o c k i t s 2-0’ ieaging software io “Voicr--ccntrolIedworkstation helps otlercoi “DO0 health systea c:st.. pp. J. pp. Hearst’s system of TileBars” (see Figure 7). on U Press. 51-58.R. up to 100 times larger than would ordinarily fit on the display. users can access pages immediately. Documents may be large and themselves have structure. Cambridge.. as in the Table Lens system. then rearrange the matrix to give clusters. “Worlds Within Worlds: Metaphors for Exploring n-dimensional Virtual Worlds. Whereas browsers currently show only a single local node. N. M. Info. Lamping. A ‘.”in Cyberspace: Fi‘rst Steps. Gershon and S. We already saw this in Figure 5. R. One of the bestknown document visualizations. they coulid be given a visualization of the infosphere with landlmarks. “Narcissus: Visualizing Information. grouping. r i c cadaver I (cosyuterii. 1995. Kodak EO1 4.V i s . Marchionini. are the elementary units of retrieval. Rao. on Human FacHierarchies. Calif.%an%’f a s t e r admissions. New York.ACM Symp. “The t l e c t . Refe’rences 1. each term being associated with a different line on tlhe strip. The figure also shows a separate inset window of code (a focus display) associated with the selected segment of the overview display. Gershon and S. S. Info. pp. 1990.19 lets users turn existing text into an indexed hypertext book with a fish-{eye table of contents. Visual representation of the matrix of correlations between word eo-occurrences or other structural parameters of text form the basis for a number of systems. J. Y “VA aL+oaation .Information visualization m x will play an important role in its success. This lets users do the most elementary operation of sensemaking. N . . pp. 76-83. L’in. (U! 7TileBars.P. 5. The WebBook8 (see Figures 1 and 6) arranges pages into a simulated physical book. 2 b i l l i o i L . “Galaxyof News: An Approach tovisdalizing and Understanding Expansive News Landscapes. Symp. SocietyJor Info. Document Finally. Calif.ACMSymp. Eick. No. 42. which.1991. The WWW is currently organized as pages. Texlt can also be msualized as long strips. they could be given avisualization of a workspace of multiple pages to compare and combine. on User Interface Software and Technology.” The Table Lens permits the display of a matrix. and each block on a strip darkens according to the frequency of those terms in that block. X. divides the text into equal blocks. Eick.. Feiner and C. eds. pp..s break $ f . Am.V 1E:EEI CS Press.ACM in UIST 90. Vol. Information access % is a “killerapp” for the 1990s. Wise et al. 3. 469. for our purposes here. instead of a single c 0 page. ihey could be given visualizations of compound E documents and other abstractions. An obvi- 66 March 1996 . Instead of waiting for each page on a hotlist to be accessed. E. VI. and P.Instead of a single page. Brushing the recency scale highlights the appropriate lines in the code text display. Hendley et al. ed. Pirolli. The matrix can be sorted by column and manipulated with other techniques to reveal patterns in the data. 4. 7. Benedikt.”J. Rennison.G. New York.” in UIST 94. Los Alamitos.G. 95.” i s . eds. Furthermore. 90-96.

and search results make interactions with DBMSs simpler and more accessible to people from all walks of life. Processing and Management. In these systems. M.318-322.. York. B.”Behavior and Info.R.”in CHI 86. in CHI94. p. 1. these systems allow for recognition and handling of complex information.” to appear in CHI 96.Nov. and E. ACM Press.E. eds. World Wide Web. 160-168. diagrams effectively display relationships among objects.pp. where direct manipulation of input devices. Robertson.ACM Press. Ahlbergand B.ACM Press. 122-128.“Pad++ :Advancesin Multiscale Interfaces. 1986. 43-54. J. many systems offer IEEE Computer Graphics and Applications 67 . 3. Bederson. Unfortunately. “Rooms:The Use of Mulfor Visualizing Software. New York. navigation. Spoerri. 1995.” in A C M tos. D Visually interacting with databases Users can retrieve information from DBMSs by issuing queries in languages such as Structured Query Language (SQL). such as sliders and toggles.“TileBars:Visualization of Term Distribution 29. 1995. Hearst. “VisualToolsfor InformationRetrieval. (2) creating a visual approach to database interfaces. Gershon et al. diffuse information.G. on Human Factors in Software.. ACM Press. 315. “Visualizationof a Document Collection: The VIBE System. Gershon and S. and icons. New York.pp. Shneiderman. Technology. Visual representations of the database structure and contents. diagrams. 1993. They also foster improved human-computer interaction and can be used by a wide spectrum of people. Proc. Software. 1996. 11. 11.” A C M Trans.pp. query. “Data Base Navigation: An 18. New York. “SeeSoft-ATool 9.K. and icons supply immediately graspable representations of object classes. New York.A. Office Environment for the Professional. Eick.1993. NO. 1987. on Human Factors in Software. and W.. Forms easily represent common properties of collections of objects. including forms.” ContextVisualization for Tabular Information. Thus. J. Los Alamitos. New York. Proc.G. 1994. 1992. July 1986. Companion. on Human Factors in Software. Interaction with Databases ramatic changes prompted by the growing global Internet affect how users deal with information. ACM Conf. Calif. N.. 55-66. 1. 175188. Symp. R. Visual query languages (VQLs) simplify formulating queries and understanding the relation between the query and query results. We are passing from a world in which a few devotees managed information to an environment in which many people consume widespread. S. Proc. 18. L. J. ACM 14.K. “Superbook IEEESymp. Remde. G. Henderson and S.”in CHI 94. Eick.”IEEE Trans. R. New York.pp. “TheWebBook 15. Olsenet al.ACM Press.Vol. “VisualizingInternet Resources. ACM Press. 211-241.pp. on 17. Along with the excitement comes the recognition that information consumers need effective and efficient tools. 12. 10. 313-317. They use visual representations of the database structure and contents. 1994. Conf. 1995. Sumner.tors in Computing Systems.”in CHI 95. Card.on VisualLanguages. 8. Query results might force the user to go over long lists of information. Calif.”Info. K. Vol. “GeneralizedFisheye Views. 19.K. Steffen. Information in Full-Text Information Access. Many systems implementing VQLs adopt different visual representations and interaction strategies. Gomez. ACM Conf.”inProc. Furnas. controls both selection of information and query formulation.G. Vol. pp.ACM Press. Los AlamiAn Automatic Tool for Information Exploration. Together with unstructured information-such as most of the documents on the World Wide Web-the information highway provides access to structured information residing in database management systems (DBMSs). access.. CS Press. Hypertext 87Proc. No. 957-968. 20. 401-408.pp. 1994.L. No.A. “TheTable Lens: MergingGraphical and Symbolic Representationin an Interactive Focus + Graphics. on Human Factors in Software. A. S. N. 95. browsing.pp.M. New York. New York. 16.”in CHI 94. and (3) interacting with hypermedia systems. 13. queries in SQL are based on math-like expressions-difficult for typical users to handle. Window-basedGraphical User Interfaces. C.pp.Vol.pp.K. ACM Conf.. Displays. even those with limited computer experience.D. on Software Engitiple Virtual Workspaces to Reduce Space Contention in neering. Landauer. 69-81. Proc. and the Web Forager: An InformationWorkspace for the ACM Conf. and T. Spence and M. and retrieval rely on (sometimes) complicated queries.IEEE CS Press. Crucial issues affect (1)interacting with databases in the global information infrastructure. We’ll explore some of them here.”inProc. Apperley. 16-23. Card.1982.“VisualInformation Seeking: Tight Couplingof Dynamic Query Filters with Starfield Press. on Human Factors in pp. 5. G. Vis. 1. Rao and S. Card. Moreover.No. ACM Conf.A.pp. IEEE Info.

Sign up to vote on this title
UsefulNot useful