Introduction

It is difficult for users to navigate large document spaces, since they often experience a disorientation problem. [3] A variety of tools are being tested to assist users. These include navigational tools such as browsers and lenses, passive processes such as history lists, bookmarks and filters, and active processes such as agents. At the University of Pittsburgh, Korfhage and Lewis are examining a variety of visual interfaces for information retrieval. [10,9,14,15] The CASCADE (Computer Augmented Support for Collaborative Authoring and Document Editing) research testbed has been developed to explore the integrated use of navigational tools, passive processes and active processes (agents) to support collaborative authoring of documents [13,7].

CASCADE employs a variety of active and passive processes that aid in the identification of important documents. CASCADE has the traditional mail filters, history lists, and bookmarks. In addition, at startup time, CASCADE produces ad hoc pages describing recent activity of users including links to relevant documents. Users can prepare "reports" with varying levels of detail on user-selected sets of documents. These reports contain system-generated links to the relevant documents. Finally, agents are under development that will identify patterns of activity on documents that may be of interest to a given user. While these active and passive processes are critical components of any integrated approach to augmented navigation, this paper reviews only those efforts to integrate a coordinated set of navigational tools to aid in the navigation and management of complex document spaces. Our work has used the standardization process as one which typifies medium-sized document spaces in which collaborative work takes place. Usage suggests that the system is also applicable to sets of legislative documents, legal cases, instructional materials, software projects, to name a few. Further, we believe the navigational mechanisms described here are applicable to any situation where complex objects with relations must be organized and navigated. Thus, for example, we believe the work is equally applicable to browsers for Web sites generally.

The current research builds on a large number of efforts, including seminal work on browsers [2,4], lens [12,11,18,1], texturemaps [16,8,15], and stucturemaps[6]. Our work looks to integrate systems of these types to provide appropriate navigational aids for the users at appropriate times. Theoretically, the work builds on Spring and Jennings [17] using the rules for mapping abstract data to virtual spaces. The research project is designed to explore the capabilities of the tools individually and more importantly in combinations that assist a user in navigation.

Each of the tools may be thought of as a map of a given scale and with a given level of detail that can be used together to aid movement through complex, abstract document space. The research makes several assumptions, including:

• A bitmap display of 1000 x 1000 pixels with a 24 bit depth and a refresh rate of 30 cycles per second may be viewed as having a bandwidth of 1,000,000 pixels * 3 bytes * 30images/second or 90MB/sec.

• Careful usage of preattentive stimuli allow users to process large amounts of data in parallel. Hue is an ideal preattentive stimulus for testing, but other less dominant preattentive stimuli may also be useful

• Navigation tasks involve at least four kinds of processes -- finding groups of objects of interest, finding specific objects of interest, following interesting paths, and tentative exploration of objects of given attributes.

• Documents and document components are objects with attributes and relations.

The tools being implemented in CASCADE attempt to capitalize on these assumptions. We view a document space as a semi-structured space populated with objects that have multiple attributes and multiple relationships. Four related tools are implemented that focus on the different kinds of navigation tasks. These include:

• Docuverse. The Docuverse tool depicts large numbers of objects. It provides an overview of the document space with the intent of helping the user to identify areas of the document space, i.e. collections of objects, that may be of interest based on recognition of specific colors, color textures, collection size, or changes in collection size. Docuverse shows minimal structure. In the prototype, this structure is the file system. The next generation will show basic document structure, but will not show full document structure as might be apparent via links. At the current time, objects represented in Docuverse show only one attribute of the selected object at a time. Thus, a given display might show the size, age, usage, number of links, or type of a document. An area of interest is selectable and selection will yield, based on the mouse button used, another iteration of a Docuverse plot with the selected area as the new center of focus, or generation of a Webview rooted on the selected area.

• Webview. The Webview tool is a traditional tree browser. The current directory (or document container) is shown with its components and containers. The containers are recursively expanded showing their contents. Individual documents and links between documents are viewed on a large pannable canvas. While the prototype currently displays only hierarchical links, future versions will display other types of links as well. Thus, instead of reflecting the hierarchical structure of a document, the Webview might show document link distances, without regard to document structure. Selecting a container object in the tree causes the root of the tree to change -- to the new container. While the number of levels shown at any time is arbitrary, we have chosen to show only three levels at the current time to maximize performance. In addition, very large directories below the first level are simply indicated and not expanded for the same reason. The user navigates up the tree by clicking on the root node of the tree. Clicking on any subnode does one of two things: 1. 1f the subnode is a directory(container) the tree is rerooted at that container; 2. if the node is a content node, it is displayed in the view window. When a document object is displayed, Landmark tools become accessible.

• Landmarks. Individual documents or components may be arbitrarily large, necessitating intradocument navigational aids. While text search has traditionally been used for intradocument navigation, a variety of additional information is becoming available with the increasing use of explicitly structured documents. Landmarks within a document may be chosen to reflect the semantic, structural or other attributes of the document. In the current implementation, two landmark maps may be appended to a document in CASCADE. Mural shows the location of all links in a document. Tilebar shows the frequency of occurrence of up to three sets of terms across a document. Each provides a view across the entire document. A variable sized thumb in the scroll bar identifies the portion of the document displayed in the view window. Navigation to any place in the document of interest is accomplished by aligning the thumb with the area of interest on the Landmark map.

• Preview. Metadata about a document can be used to provide a lens with information for the user about the desirability of opening a given document or traversing a given link. Most lenses attempt to assist the user by providing information about a given document in advance of a commitment by the user to traverse a link or open a document. In CASCADE, there are two kinds of previews. First, all anchors are color-coded to define the type of the destination node, as they are in the Webview. Thus, visually, the user knows the type of node that is at the end of the link simply by glancing at the color. The anchor also carries the name of the destination node, and for system-generated links, this identifies the author and the creation date. The second level of preview is accomplished by clicking on the anchor which brings up a small window which provides information about all the attributes of system-generated nodes -- for the purpose of the standards CASCADE research, these kinds of nodes were generated when users commented on a document.

The next four sections describe these navigation tools in somewhat more detail.