User Guide for the NIST Information Retrieval Visualization Engine (NIRVE)

User Guide for NIRVE <-- You are here

1. Overview

This document gives detailed instructions on how to use the various features of NIRVE. For a more abstract discussion of rationale and purpose, see Document Clustering in Concept Space. Some of the details in this paper are out of date, but the basic ideas have remained the same.

The purpose of NIRVE is to allow you to visualize and manipulate a set of documents resulting from a query to a search engine. The query is characterized by a set of keywords. NIRVE will let you consolidate these keywords into a smaller set of concepts and then will organize the documents into groups, called clusters, based on the documents' concept profiles. You can inspect individual documents and clusters. You can also suppress viewing of selected documents and clusters so as to concentrate on items of greater interest.

NIRVE displays two windows to the user. The main body of this document explains the appearance of, and operations within, each of these windows. Here are two screen dumps illustrating the general appearance of these two windows, one with a more global view, the other with a detailed view.

NIRVE Control
NIRVE Control is a "menu" window used for overall control and mode-setting. Information about the various options can be found during execution by clicking on the little question-head at the left end of each line. Operations include database query, concept control, opening and closing clusters, filtering results, viewing, and finally summarization, and quit.
Document Space
Document Space contains the main display of icons representing the selected documents, and clusters thereof. You can see this display from any desired viewpoint, and request titles and/or full text of any document. Also, you can assign a value to documents and clusters representing your judgment about their usefulness to you.
Concept Control
Along the bottom of the Document Space window are arrayed the grouping of keywords into concepts that control document clustering. Concepts are color coded, and the cluster icons display a color histogram representing the frequency of each concept within the cluster.

2. Installation

NIRVE is composed of several windows, files and processes. Here is an system overview of the way various components interact. In order to do new queries, NIRVE must be hooked up to the PRISE software in the appropriate way. A description of PRISE is beyond the scope of this user-guide, but may be obtained from the WWW. Here are the basic steps you will need to go through to build NIRVE from source code on your system:

2.1 Software support

You will need at least:

2.2 PRISE database

You must have a PRISE database index of document titles, for retrieval based on keyword. Specifically, you will need the files conventionally named: postings, titles, titles_tables, docstats, and tcollstats. The program extract uses the first three to perform retrievals and make_lengths uses the last two to generate the doc_lengths file. Note that postings can be rather large, typically occupying tens or even hundreds of megabytes.

2.3 Copy of source code

All the source code and related information may be found at the NIRVE FTP site.

3. Invocation

You can control the operation of NIRVE through the use of command-line options and environmental variables. Default values are in effect for any unspecified values.

3.1 Options

-sb
Enables handling of spaceball events. If omitted, the default is to ignore such events.
-zrot
Allows rotation of the icon array around the z-axis (sideways tilting). If omitted, the default is to disable such rotation.
-sync
Recommended for slower machines; keeps mouse input and screen appearance in synchronization, but may hurt performance. The default is not to force synchronization.
-flat
Recommended for machines with only an 8-bit color plane; causes NIRVE to use flat lighting, instead of reflective. May improve performance, although appearance may be slightly degraded. The default is to use reflective lighting.
-density <real-value>
<real-value> is a measure of how closely the icons are packed together. The default is 0.5. A higher value implies closer packing.
-mcpr <integer-value>
<integer-value> determines the maximum number of concepts per row of the legend at the bottom of the docspace window.

3.2 Environmental Variables

If you wish to change the defaults, these variables would normally be specified with the Unix setenv command.
NIRVE_TCL
This is the name of the Tcl script that generates the NIRVE Control window. The default value is "./sph-ctl.tk".
CHAR_GEOM
This is the file containing geometric descriptions of the ASCII characters. The default value is "/home/cugini/grfx/char-geom.dat".
Back to NIRVE Home Page