The Category Map which is output from cMap is then graphically displayed in a three-dimensions to allow the user to "fly through" the clusters in an effort to explore and browse the dataset.

This Information Spaceflight will enable users to browse a large, unknown dataset. The user will quickly be able to zoom in on an area of interest and subsequently search that area with more traditional approaches.

We use a custom implementation of the SOM algorithm, which we call cMap, to cluster and categorize large datasets of documents. This custom implementation has a number of optimizations over the basic SOM algorithm; these optimizations have improved the run-time of the algorithm by many orders of magnitude. cMap can take an original multidimensional dataset of documents and create a two dimensional data space, or "map", of the document clusters. By examining the distribution of the documents in this map, we are able to create a three dimensional landscape visualization which can easily show the relationships and degree of similarity between the document clusters.

Our current Interspace Prototype consists of semantic indexes for several subject domains and an analysis environment that enables concept navigation within and across these domains. The server side uses the Versant object-oriented database to store the indexes, while the client side uses the ParcPlace Smalltalk user environment to support the navigation. This web page uses the ClassicBlend software to transform the Smalltalk screen displays on-the-fly into Java screen displays so that the full functionality can be invoked from within a web browser (with a text-only interface that is more awkward than the graphics-oriented Smalltalk one — screendump of the Smalltalk interface).

The current domains are Colon Cancer and Osteo Arthritis (30K and 10K abstracts respectively each in clinical medicine from MEDLINE), Artificial Intelligence (20K abstracts in computer science from COMPENDEX), and Asian Economics (15K abstracts on business manufacturing from ABI/Inform).

To invoke the IRA (Interspace Remote Access) from a web browser, you need the latest Java-enabled browser, i.e. Microsoft Internet Explorer 4.0 or Netscape Communicator 4.5 . Pressing the TRY DEMO button starts the client. You will be asked by Java for permission to invoke ClassicBlend — please click Grant this permission. Then wait patiently and the initial window will start up. A separate program has now been invoked and you are interacting with the multiple window environment of the Interspace Prototype.

The main window has 3 panes: domain (left), search (upper right), concept (lower right). The left button selects and the right button when held down brings up the menu. The domain pane starts with the list of collections available, and can be used to select which subdomains (categories) to navigate. The search pane enables a desired concept phrase to be specified. The current domain or subdomain is then searched, and the list of concepts appear in the concept pane. At any time, the menu can be used to display related concepts or related documents. If a document is selected, it is displayed in a separate window.

Using category maps and concept spaces, the Interspace can be navigated.

To navigate using category maps to get the lay of the land for the collection. Select a domain and issue subdomains. Search the subdomain for a desired phrase. Look through the concepts and display related documents for relevant ones. Select a desired document and display it. From the document, the related concepts are displayed and the navigation can continue.

To search using concept spaces to locate desired terms or documents. When concepts are displayed and the list is not satisfactory, select terms that have an exclamation point (!) in front. A list of related terms within the collection will be displayed and navigation can continue.

For more information see: Using The Interspace Remote Access Client.