Document Space


User Guide for NIRVE

Overview

The Document Space window is the main area for viewing an abstract representation of a selected set of documents. The documents are arranged in groups called clusters, based on similarity of their concept profiles. Each cluster has an icon that displays the average concept profile for the documents therein and a user-assigned value. The concept profile is displayed via a histogram whose bars are color-coded to match the colors of the concept labels along the bottom of the screen.

Cluster icons are arranged on the surface of a globe. Their latitude is determined by the number of concepts within the cluster: more concepts cause an icon to be located nearer to the 'North Pole' of the globe. Longitude has no intrinsic meaning; icons are arranged so as to try to place clusters with similar concept profiles near each other. Clusters which differ by a single concept are connected by an arc whose color represents the conceptual difference between them. E.g. if cluster A has the concepts 'boat', 'sink', and 'ocean', and cluster B has 'boat', 'sink', 'ocean', and 'storm', then they will be connected by an arc color-coded for 'storm'. The thickness of the cluster icon box is proportional to the number of documents contained therein.

When a cluster is opened, a 2D rectangle is projected outward from the cluster icon. This rectangle contains all the document titles within the cluster. These titles are arranged such that similar titles (i.e. those containing some matching words) have nearby horizontal positions. Vertical position is controlled by the score assigned by the search engine. Thus similar titles will appear in the same column, with better scores towards the top of the column.

Clusters from concepts

A document's concept profile is simply the set of its strengths for each of the active concepts. Individual concept strength for a document is calculated based on the frequency of occurrence of the keywords that have been mapped into the concept, and on the document length. More occurrences imply greater strength, as does shorter length (since that implies a higher frequency rate). Document concept profiles are not displayed directly, but are used to sort documents into clusters.

By definition, two documents are in the same cluster iff they have non-zero values for the same set of concepts. Thus, if there are five active concepts, there are at most 32 clusters. The basic information displayed for each cluster icon is a graphical representation of its concept profile (representing the average value for the documents contained therein). This is displayed as a colored histogram, one bar per concept. The correspondence between color and concept is shown in the concept legend at the bottom of the window.

Operations in Document Space

The window is always in one of three view modes: spin, move or pick. Different operations are available in each of these modes. The following table presents an overview of the three modes.
 Mode  |  cursor           MB1                 MB2         MB3
 ------+--------------------------------------------------------------
 spin  |  circle           --                  --          --
 move  |  cross-arrows     rotate lat/long     shift xy    shift z
 pick  |  pointing-arrow   show icon contents  mark icon   position icon
You may use the NIRVE Control menu to switch among view modes. The current view mode is represented by the state of the radio buttons in the view mode entry of the menu, as well as by the cursor in the document space window.

You may also use shortcut keys to accomplish the same effect. (A shortcut key is a single keystroke typed at the document space window.) If you are in pick mode, the space bar switches to move mode; if you are in any other mode, the space bar switches to pick mode. Generally the space bar is used as a toggle switch between move and pick mode. If you are not already in spin mode, the 's' key switches to spin mode. When the spaceball is enabled, its use switches NIRVE into move mode.

Spin Mode

Spin mode is denoted by a circular cursor in the document space window. The mouse is inactive. The document space display is rotated at a constant rate about its natural axis. You can adjust the speed and direction of spin via the appropriate entry in the NIRVE control window or you may use shortcut keys as follows: Note that within spin mode, the 's' key means "spin more slowly", but otherwise it means "switch to spin mode".

One recommended use for spin mode is to scan through the cluster icons to get a general idea of which concepts are represented (and which not) among the clusters.

Move Mode

Move mode is denoted by a cross-arrow cursor in the document space window. The mouse is used to control the position of the array of document and cluster icons. Holding down mouse button 1 (MB1) allows the user to rotate the display, with horizontal motion controlling longitude and vertical motion controlling latitude. MB2 controls translation of the display horizontally or vertically. Finally, vertical movement with MB3 held down controls translation of the display towards or away from the viewer (i.e. along the "z" axis).

When the spaceball is enabled, it also may be used to position the document array. Rotation around the x-axis (latitude), the y-axis (longitude) and 3D translation may be performed simultaneously. Note that rotation around the z-axis (sideways tilting) is normally disabled for the spaceball as for the mouse.

Pick Mode

Pick mode is denoted by a pointing-arrow cursor in the document space window. The mouse and/or shortcut keys are used to inspect and mark documents or clusters. The following table summarizes the pick mode operations.
                 |  mouse action   shortcut keys
-----------------+----------------------------------------
icon selection   |  motion
show contents    |  MB1            t    
mark icon        |  MB2            b/u/g
icon close-up    |  MB3            n/p (for next/previous) 

Moving the cursor over an icon (with no buttons pressed) causes the icon to be highlighted. Cluster information is displayed at the top of the window. When an icon is highlighted, you can use the shortcut keys 'n' and 'p' to highlight the next or previous document or cluster and to position it for detailed viewing. This may be more convenient for systematically scanning through a sequence of entities than sliding the mouse.

Pressing MB1 on a document will display the full document text, using Netscape. Within the document, keywords will be highlighted in the color corresponding to the concept containing that keyword. Here is an example of a formatted document. Pressing MB1 on a cluster icon acts as a toggle switch to open or close the cluster. Opening the cluster generates a display of the contained document titles as discussed above in Overview. The same effect is achieved on a highlighted icon by the shortcut key 't' (for text).

Pressing MB2 on a document or cluster icon will cycle through the user relevance values for that item, denoted by a small flag attached to the upper right of the icon: red for bad, yellow for unsure, green for good. These values may also be assigned by the shortcut keys 'b', 'u', or 'g' respectively. Note that when assigning a value at the cluster level, all documents within that cluster inherit the value. If you then cause a re-clustering (e.g. by changing the keyword to concept mapping), the newly generated clusters are assigned a value based on the values of their constituent documents: if all the documents are good, the cluster is good, if all bad, the cluster is bad, otherwise the cluster is marked unsure. Documents are persistent entities, but clusters are comparatively transient: they are (re-)generated as needed during the session.

Pressing MB3 on an icon will swing it around front and center for closer viewing, similar to the effect of the 'n' and 'p' keys.