Taking a step back and reviewing all of the principles discussed so far, we can see how they combine to provide a set of conceptual tools for designing navigable information spaces (Table 9.1).
They seem diverse in intent and scope, but together provide a set of preliminary guidelines to design for information navigation. They are best-suited for spaces that communicate by providing a meaningful path through the content of the space. A space that lets the navigator move as quickly as possible to information of interest would have a different character, and require a different set of design principles. Put another way, a library is organized very differently than an educational exhibit. Even so, some of the principles would still apply in a space designed for efficient searching, where there are many navigation choices to make and design for effective wayfinding becomes paramount.
The field of information visualization seeks to graphically depict existing document structure. Relevant to this thesis are visualizations that employ a spatial metaphor, or interaction based explicitly on navigation. The Galaxy of News system constructs a three-dimensional information space of words from news stories that users can fly through [Rennison, 1994]. The VIBE system places document icons in two dimensions based on term-vector similarity to pre-stored queries [Olsen et al., 1993]. The Harmony VRWeb browser embeds hyperlinks into a three-dimensional model of the city of Graz, Austria [Andrews, 1995]. Xiong is developing a system to transform Web sites into navigable, immersive environments that support social interaction among browsers [Xiong, 1997]. Cone trees, three-dimensional visualizations of tree structure [Robertson et al., 1991], have been used for Web sites [Carrière and Kazman, 1995], a large medical article classification scheme [Hearst and Karadi, 1997], and other hierarchies. Information visualization focuses on presenting existing structures spatially, as opposed to designing a space that incorporates a novel organizing principle. For a survey of recent work in information visualization, see [Card, 1996].
Some researchers have designed information spaces to explicitly represent and support cognitive mapping. Trajectory mapping (TM) [Richards and Koenderink, 1994] is a technique that combines subjects' associations among subsets of information items to construct a path through them. This technique was applied to visualize an information space of tourist attractions in Boston [Lokuge et al., 1996]. Other work has focused on making an existing information landscape more imageable [Chalmers et al., 1996,Chalmers, 1993]. In this work, an existing visualization of research articles was augmented with static imageability features such as paths and regions, and dynamic features that enlarge document titles according to user focus.
Furnas describes information navigation in terms of effective view traversibility (EVT) and effective view navigability (EVN) [Furnas, 1997]. An information space that is effectively view traversible maintains relatively short path lengths as the number of nodes or views grows. For example, a hierarchy is efficiently view-traversible because the longest path in a hierarchy with branching factor b remains O(logb n) as the number of nodes nincreases.
Efficient view navigability requires that the navigator can find his way to a goal in the space. To do so, he examines the local ``residue'' of the goal in the current view - information that would lead him to move closer to it. An example of such residue is the label of a node in a hierarchical classification, which indicates an attribute shared by all of the items in its subtree. The principle that states ``provide signs at decision points'' can be reinterpreted as ensuring sufficient local residue to make the information space effectively view navigable.
Hypertext researchers have argued for more effective spatial representions of hypertext for some time [Dieberger, 1997,Masuda et al., 1994]. Kaplan and Moulthrop argue that attempts to map the n-dimensional ``semantic space'' of a hypertext onto two- or three-dimensional representations will always be imperfect [Kaplan and Moulthrop, 1994]. Hypertext representations carry over the underlying link-node structure of the hypertext model and its visibility and mappability constraints.
Finally, Zellweger introduces the notion of scripted documents, which explicitly support authored paths in a hypermedia system [Zellweger, 1990]. The path scripts could specify sequential paths, in which the viewer sees a preset sequence of documents; branching paths, in which the viewer could make a choice between one of two paths; and conditional paths, in which the next document presented is conditional on some author-specified test.
One topic that has not been addressed in depth or generality is process, a procedure for taking a collection of information and applying these principles to generate an information space. The ordering of the presentation of the principles is suggestive of the notion that conceptual organization should precede spatial organization, and the computational medium used to enhance the space whenever possible. In a given situation, though, the principles can interact nonlinearly, with the application of one impinging on the requirements of another and forcing tradeoffs.
Those considerations granted and using what we have learned from the two spaces described here, we can make a brief sketch of a process for designing spaces that communicate:
A process will also have to account for the fact that information is inherently dynamic. Its veracity and relevance to an audience changes over time. This is reflected in the need to augment the JAIR information space with articles as they are published, and to add and remove subject listings from the Course VI information space for each term. Any implementation of an information space will need to address issues of dynamism and maintenance.
We would also like to have tools that can automate the application of any of the principles presented above. A system that can take a database or hypertext and delineate useful landmarks, paths and regions, or trace out conceptual dependencies (even in an approximate fashion) would be of great use in sketching out an information space. The capabilities of such a tool would depend crucially on how well-structured and annotated the original source of information is, however.
The greatest benefit, though, could come from capturing an information space as a solution to a design problem in a way that could be applied to similar problems. Perhaps a relatively small number of effective clichés of spatial organization, for communication, search, or another task, are all that is needed to account for a large number of these solutions. Such clichés are exemplified by libraries, bookstores, and trade-show exhibits, in addition to the educational exhibits. In this case, another thrust of inquiry would be to identify and generalize these clichés as sources of knowledge for design by analogy. Process, then, could become the application and specialization of the appropriate cliché to the information collection at hand.
We began by presenting information-seeking as navigation through an information space, and by asking how we can structure that space to facilitate a particular information-seeking task. This motivated the study of educational museum exhibits, exemplars of physical information spaces that communicate effectively to their visitors by allowing them to navigate through knowledge. The result is a collection of principles, which are guidelines for taking a collection of information, organizing it spatially, and presenting it to an audience.
Two questions about those principles that can be answered by the experience of designing more spaces are,
To make our list of principles more complete and effective, we can also look to other physical spatial organizations of information. For example, libraries can provide a rich source of knowledge about building spaces for both browsing and retrieval. Study of the process of building an effective library classification system for subject browsing could provide principles for spaces that are used more like libraries.
And finally, we can evaluate how the spaces are used, and whether the users can solve useful tasks in them. In a space that communicates, we can test what knowledge the user has gained by navigating through the space, as well as noting any wayfinding difficulties the user experiences in the space.
All of these steps take us closer toward a collection of design principles that can help users navigate through the increasing volume of information available in electronic form. It is hoped that a eventually a sufficient set of principles can be found that will bring clarity and insight to a broad conceptual framework for organizing information for navigability.